|
A peer-reviewed electronic journal published by
the Institute for Ethics and ISSN 1541-0099 28(1) – September 2018 |
The Value of Consciousness and Free Will in a Technological
Dystopia
Allan
McCay
University
of Sydney Law School
University
of Sydney Foundation Program
Centre
for Agency, Values, and Ethics, Macquarie University
Journal of Evolution and Technology - Vol. 28
Issue 1 – September 2018 - pgs 18–30
Abstract
Yuval
Noah Harari warns of a very pessimistic future for our species: essentially,
that it may be superseded by non-conscious Artificial Intelligence that can do
anything we can and more. This assumes that we are physically instantiated
algorithms that can be improved on in all respects. On such an assumption, our
labor will become economically worthless once AI reaches a certain level. This
picture, however, changes markedly if we accept the views of David Hodgson in
respect of consciousness, free will, what he calls plausible reasoning, and the
relationship among these. On Hodgson’s account, there will always be valuable
skills requiring a particular kind of judgment that are possessed by humans,
but not by non-conscious algorithmic machines, however advanced.
Introduction
Are
consciousness and free will worth anything? In this paper, I will focus on the value of consciousness and free will within
economic systems of the future. I will do this by considering a recent book
that envisages some dystopian future scenarios, in light of work at the
intersection of philosophy of mind and the free will debate. The intention is
to thereby consider philosophical debates about consciousness and free will, in
a context that is some distance from traditional treatments of these topics –
my context being one that is primarily concerned with the economic worth of human labor in a technological dystopia. In
particular, I focus on the question of how humans will fare in economic
competition with Artificial Intelligence (AI).
Homo Deus: A Brief History of Tomorrow (2016) is a
popular book (written by Yuval Noah Harari, an academic historian1 specializing
in human history on a grand time scale) that is having a significant impact on
popular debates about the future of technology and the future of humanity.
According to Harari, we face a bleak future, and may become economically useless,
as a result of being superseded by non-conscious AI and/or enhanced humans.
In
developing his thesis, Harari attaches great significance to the concept of the
algorithm. Throughout the book he works with a presupposition about the role of
algorithms in human behavior that he claims is becoming a dogma2 in
scientific discussions and has also much wider impact on society. The
presupposition is that humans are algorithmic choosers who lack free will.
Whilst he seems concerned about this perspective on agency and regards it as
important to expose it for critical evaluation (Harari 2016, 397), he also
appears to find it plausible enough for it to be the view that he assumes in
relation to his predictions about the economy. Harari’s assumption about algorithmic
human agency has great import for the way he envisages that we will engage in
economic competition with AI and, in short, are likely to be outperformed. As a
result of the book’s focus on the relative merits of humans and AI in the
workplace, Harari’s argument provides a useful framework in which to consider
an issue that is of great contemporary significance – that of the automation of
work.
Whilst
David Chalmers, for one, has considered the possibility of non-algorithmic AI
(2015, 177), it is significant that contemporary
approaches to AI are algorithmic in nature. Although recently there has
been substantial progress in AI, for example in the branch of machine learning
known as artificial neural networks (which is loosely inspired by the neural
structure of the brain), it is important to note that, like earlier forms of
AI, neural networks operate algorithmically. Although the algorithms that
constitute a neural network can facilitate the machine learning required for
impressive applications such as self-driving cars, the code that underpins such
learning is still algorithmic.
For
Harari, we have something in common with self-driving cars and other forms of
AI insofar as the way we make decisions and navigate the world is also
algorithmic. According to Harari, the risk to our economic value comes from a
trajectory in the development of technology in which the algorithms
underpinning AI outperform our more modest algorithmic agency in all ways that
are economically useful.
However
the philosophical work of David Hodgson challenges the analysis presented by
Harari by suggesting a different picture of human agency. Hodgson also
considers the role of algorithms, focusing on human decision-making, but in
contrast to what is assumed in Harari’s book Hodgson argues that human
decisions are not merely algorithmic
choices, but involve “plausible reasoning,” which, according to him, gives a
role for consciousness and a basis for free will.
As
with Harari, David Hodgson (1939–2012) has an unusual background as a
contributor to debates about free will and consciousness. After completing
doctoral studies under the legal philosopher H.L.A. Hart, Hodgson went on to
become a Judge of Appeal in the Supreme Court of New South Wales, Australia,
and whilst on the bench he published philosophical work “at a rate that would
be respectable for a full-time philosopher” (Levy 2013). In his later years,
most of his work focused on free will and consciousness, culminating in the
publication of his third and final monograph, Rationality + Consciousness = Free Will (2012) in which he argues
for a libertarian form of free will that, he claims, could make people morally
responsible and deserving of punishment for what they do. It is the view of free will and consciousness in this last book
that I will pit against Harari’s analysis, and I will consider the implications
of accepting Hodgson’s view for Harari’s pessimistic predictions.
Why
consider Hodgson’s work in response to Harari? There are three reasons to focus
on Hodgson. The first relates to Hodgson’s discussion of algorithmic choice.
Within the free will debate, Hodgson is perhaps the writer who engages in the
most substantial discussion of algorithmic choice, and his comments on this
issue raise questions about Harari’s thesis. From time to time Hodgson compares3
algorithmic and non-algorithmic agency, thereby providing useful insight for
the analysis in this paper.
A
second reason relates to Hodgson’s focus on consciousness. One of the main
questions in Harari’s book, and this paper, relates to the value of
consciousness in a technological dystopia. As noted by Levy (2013, 185), a
significant feature of Hodgson’s view is his emphasis on the role of
consciousness in free will. Arguably, Hodgson’s work has the most substantial
discussion of consciousness in contemporary debates about free will (aside from
the empirical perspectives discussed by Levy (2017)).
A
third reason relates to Hodgson’s judicial background. Hodgson uses examples
from his work as a judge to illustrate the kind of task for which consciousness
and non-algorithmic reasoning are important. As Harari’s thesis engages with
the future of employment, it is useful to consider Hodgson’s vocational
(judicial) examples.
Although
Nagel (2012) has described Hodgson’s last book as a “persuasive account of the
relationship between consciousness and free will,” critics of the work have
drawn attention to issues with various aspects of Hodgson’s argument, ranging
from those relating to his perspective on the laws of nature, to his analysis of evolution’s role in the
emergence of consciousness, and the ability of his account of free will to
address concerns about control and luck, and to form a basis for retributive
punishment (Carruthers 2012; Capes 2012; Earp 2013; Levy 2013; White 2013;
Deery 2015; de Sio and Vincent 2015).4 It is not possible to respond
to these issues in a paper which addresses Harari’s book, but it is worth
noting that many of the criticisms do not relate to the economic argument presented here. For example, it is not necessary
for me to assume that humans have a form of free will that might make them
deserving of retributive punishment in order to show that their agency provides
them with a capacity that is of value in a labor market.
My
goal here is thus not to defend Hodgson’s theory but to raise a serious
possibility that what he calls plausible reasoning may give humans an economic
advantage over non-conscious AI, and to suggest that this view should be borne
in mind when considering the basis on which Harari’s predictions are made, and
the likelihood that they will come to fruition. I thus assume that Hodgson’s
views (or something like them) are defensible against pertinent criticisms,
then apply them in the consideration
of claims about future economic circumstances.
I
begin by briefly saying something about the aims of Harari’s book before
setting out his claims about algorithms, consciousness, and free will. Later I
outline various future scenarios that are developed in reliance on the claims.
After that I use Hodgson’s work to draw the claims into question, before
revisiting the grim economic scenarios that Harari contemplates.
My
Hodgson-inspired response to Harari suggests that, in the technological
dystopias that Harari envisages, libertarian agents who can use a reasoning
facility that derives from their consciousness might have a superior economic
value to deterministic agents (at least of the algorithmic variety).
Libertarian free will and consciousness may thus have a dollar value, and if we
are indeed libertarian agents of the
type described by Hodgson, our economic future might be less bleak than if we
were deterministic agents, whose method of choosing is governed by rules.
Thus
the following argument is methodologically unusual insofar as it challenges
claims about the economy using a theory of free will and consciousness. This
challenge to an economic forecast constitutes a novel application of such
philosophical theorizing.
Harari
on algorithms, consciousness, free will,
and dystopia
In Homo Deus: A Brief History of Tomorrow, Harari
sets out a number of possible future scenarios that might result from
scientific and technological progress. He does not claim that any scenario is
inevitable, and the book aims to identify some broad trends in the development
of human societies in order to consider the merits of progressing one way or
another. As a historian concerned with the macro trends in the trajectory of
humans, Harari develops an analysis that is informed by a significant
engagement with the past.
Harari’s
book could be said be to a contribution to the transhumanist tradition,5
which considers ways in which the human condition might be transformed in the
future. He claims that humans now have the capacity to control famine, plague,
and war (even though this capacity is not always exercised due to political
failure). He anticipates that, having attained the ability (albeit not always
the motivation) to control such phenomena, humans will turn their minds to the
pursuit of immortality, happiness, and divinity (in the form of great powers)
(Harari 2016, 21). According to Harari, the pursuit of these aims will be by
way of biological engineering, cyborg engineering, and the creation of AI
(2016, 43). He expects great technological advance in these areas.
This
paper focuses primarily on the economic threat from AI, and in this connection
it is worth noting that Harari envisages impressive forms of AI that lack
consciousness.6 He does not address the functionalist7
account of consciousness, and his dismissal of conscious AI is not fully
supported. However, he says, “[o]ver the past half-century there has been an
immense advance in computer intelligence, but there has been exactly zero
advance in computer consciousness. As far as we know, computers in 2016 are no
more conscious than their prototypes in the 1950s” (Harari 2016, 311).
From
this he appears to assume that “it doesn’t seem that computers are about to gain consciousness” (Harari
2016, 311; emphasis added), and, although he does not specify a time frame for
the emergence of the scenarios he warns of, his book is best understood as
referring to a time in which AI is not conscious. He does not seem to suggest
that AI will never become conscious, or that it will at some point, but he does
not treat conscious AI as a serious possibility in respect of the economic scenarios
he considers.
This
is an important point for my purposes, as my focus is on the economic
advantages of free will and consciousness over non-conscious AI and my argument only relates to a time when AI
is not conscious. I am agnostic on the question of whether AI will at some
point attain consciousness, and I will return to the possibility of conscious
AI briefly toward the end of the paper.
Prior
to considering the economic scenarios that Harari sketches, I will say
something about the way that Harari conceives of human agency.
Algorithms
and free will
As
will be seen in the next section, an important premise in Harari’s theorizing
is that humans are algorithmic choosers who lack free will. This view of agency
becomes the model of workforce agency that he assumes of humans when
considering the relative values of humans and AI in the labor markets of the
future. For Harari, algorithms are becoming “the single most important concept
in our world” (Harari 2016, 83). Because of the importance of the concept for
Harari (and for this paper), it is worth setting out how he defines it:
An algorithm is a methodical set of steps that can be
used to make calculations, resolve problems and reach decisions. An algorithm
isn’t a particular calculation, but the method followed when making the
calculation. (Harari 2016, 83)
Why
are algorithms so important? AI depends on algorithmic methods, by way of
codified procedures, but another very important reason for the significance of
algorithms is that, according to Harari, biologists have now concluded that human beings are also algorithms
(Harari 2016, 85). He elaborates on this idea with a comparison between humans
and coffee vending machines which, in algorithmic fashion, go though a series
of steps to produce a cup of coffee. Harari sees humans as vastly more complex
algorithmic agents: “The algorithms controlling vending machines work through
mechanical gears and electric circuits. The algorithms controlling humans work
through sensations, emotions and thoughts” (2016, 85).
In
the analysis presented by Harari, “free will exists only in the imaginary
stories we humans have invented” (Harari 2016, 283), and the algorithmic nature
of decision-making has a role in our unfreedom. He says, therefore: “The algorithms constituting
a human are not free. They are shaped by genes and environmental pressures, and
take decisions either deterministically or randomly – but not freely” (Harari
2016, 328).
For
the purposes of the discussion of Hodgson’s work, it is useful to briefly
contextualize Harari’s comments on free will within the views prevalent in the
scholarly debate. Much of contemporary philosophical debate about free will is
focused on the compatibility of free will, moral responsibility, and/or
blameworthiness with an indeterministic or deterministic picture of the world.
Within that vast debate, some are compatibilists. These are philosophers who
think that at least some of the following concepts are compatible with
determinism: praise, blame, and deserved reward or punishment. Other
philosophers think that concepts such as these presuppose libertarian free
will. According to libertarians in the free will debate, praise, blame, and
desert presuppose an undetermined
form of choice, which is not random but controlled by the agent.
Whilst
Harari is not primarily concerned with blame and punishment, what he refers to as
free will is libertarian free will.
Harari conceives of freely willed decisions as being “neither deterministic nor
random” (Harari 2016, 281). For the purposes of his theorizing, Harari assumes
that humans do in fact lack free will, and he suggests that scientific and
technological advance means that people in general will lose their belief in
free will:
Doubting free will is not just a philosophical exercise.
It has practical implications. If organisms indeed lack free will, it implies
that we can manipulate and even control their desires using drugs, genetic
engineering or direct brain stimulation. (Harari 2016, 286)
Thus,
according to the view of agency that Harari works with, we are algorithmic
choosers who lack free will, and who may be subject to various forms of
manipulation based on new understandings of our unfree decision-making
processes. But how does this view of agency figure in Harari’s predictive
economics?
Dystopias
Harari
identifies three possible dystopian scenarios that might emerge from
technological advance. One sees members of the species Homo sapiens completely losing their economic value, another involves
them becoming managed by AI, and the third envisages the rise of
technologically enhanced superhumans.
In
describing the first scenario, Harari argues that the link between intelligence
and consciousness is breaking (Harari 2016, 311). Humans have general
intelligence and consciousness, but he envisages the rise of AI that is not
conscious but may still outperform us. The significance of algorithms to
Harari’s thinking emerges in the following passage:
The idea that humans will always have a unique ability
beyond the reach of non-conscious algorithms is just wishful thinking. The
current scientific answer to this pipe dream can be summarised in three simple
principles:
1. Organisms are algorithms. Every animal
– including Homo sapiens – is an
assemblage of organic algorithms shaped by natural selection over millions of
years of evolution.
2. Algorithmic calculations are not
affected by the materials from which the calculator is built. Whether an abacus
is made of wood, iron, or plastic, two plus two beads equals four beads.
3. Hence there is no reason to think that
organic substrates can do things that non-organic algorithms will never be able
to replicate or surpass. (Harari 2016, 319)
Thus,
Harari assumes that we are algorithms, our organic substrate is not
significant, and there is no reason to think that an algorithm lacking in
consciousness, and operating on a non-organic substrate, will not be able
operate in the same way as us or better.8 Thus non-conscious AI may
one day equal or surpass us in intelligence. For Harari, the full import of
this claim comes from a consideration of the effects on employment. Because AI
will equal or surpass us, it will make us economically useless.
In
relation to the second scenario, management of humans by algorithms, Harari
claims that artificially intelligent algorithms will have the capacity to
understand9 us better than we understand ourselves. Non-organic
algorithms, such as those provided by large technology companies, may come to
monitor our bodies as we interact with the world, learning about what we like
and dislike, and what stresses us or relaxes us. They may also learn our
interests by monitoring what we read.
Accordingly
many people will transfer authority to the external algorithms, and let them
make decisions. People will no longer see themselves as being autonomous and in
charge of their lives (Harari 2016, 329). Even decisions such as choice of
partner will be made by artificially intelligent algorithms, since they will be
regarded as more competent (Harari 2016, 337). The upshot of this transference
of authority is that that humans might ultimately be managed by a large network
of algorithms. Thus, even if humans retain some usefulness within the system,
they will defer to these artificially intelligent algorithmic choosers.
The
third scenario Harari envisages involves a split in humanity, in which some
humans are upgraded through technological progress, and they make the
significant decisions for society, whereas other humans are subordinate to
these superhumans (and also to non-human algorithms) and are regarded as
useless.
But
are we algorithms?
Harari
concludes by posing a series of questions, which he says he hopes will engage
the reader long after finishing the book. One of the questions he asks is “are
organisms really just algorithms?” (Harari 2016, 397). This important question,
which will be addressed in the next part of this paper, is of great
significance to his thesis. Indeed, earlier in the book he acknowledges that
computers might have inherent limitations, compared to organisms, if the latter
are not algorithmic: “If organisms function
in an inherently different way to algorithms – then computers may work wonders
in other fields, but they will not be able to understand us and direct our
life, and they will certainly be incapable of merging with us” (Harari 2016,
345).
Before challenging
Harari’s predictive project, I will state the premises used in the development
of his thesis that are to be questioned. These are as follows:
1. Human
choice is algorithmic
2. Humans
do not have libertarian free will
Challenging
the algorithmic claim: Hodgson on consciousness and free will
If
Hodgson is right then both of these premises are false. Human choice is not algorithmic and we have libertarian free will. According to him, it is our
consciousness, together with a form of rationality that is unavailable to
non-conscious AI, that gives us libertarian free will. Acceptance of Hodgson’s
view would lead to a rejection of some of Harari's assumptions about the agency
of workers, and it would show a need to reconsider the scenarios envisaged.
Before considering its
implications for Harari’s predictions, I will summarize Hodgson’s account of
free will. There are a number of steps in the development of the theory and it
is not easy to condense it into short form, but I will now give an outline
prior to applying it in the next section.
Hodgson claims there are
two types of reasoning that humans can perform: algorithmic reasoning and
plausible reasoning. According to Hodgson, humans are able to engage in formal
reasoning (such reasoning takes place in logic and mathematics). In these domains,
the conclusions are entailed by the premises and rules. Hodgson also calls this
algorithmic reasoning, and there is a mechanical sense to such reasoning which
allows computers to operate this way. Hodgson and Harari could agree that
computers work algorithmically, but importantly, whilst Hodgson would say that
humans can operate algorithmically, contrary to the view of human agency
assumed by Harari, Hodgson does not suggest that this is the only way that humans operate.
According to Hodgson,
humans also engage in reasoning in which the conclusions are not entailed by
the reasoning process, because the reasons are inconclusive. This form of
reasoning, he calls plausible reasoning (Hodgson 2012, 37–53).
A form of plausible
reasoning takes place when humans are weighing incommensurables. Thus if one is
deciding whether to honor a promise to help a friend, or to go for dinner with
a person one finds attractive, the competing reasons are of a different kind
(duty and desire) and thus are incommensurable. The reasons do not lead to
conclusions through entailment, and judgment is required. This involves
reasoning about what to do, but
Hodgson argues that reasoning about what to believe
is also a form of plausible reasoning. An example of this comes from Hodgson’s
judicial experience in which he describes a judge trying to decide what to
believe after hearing inconclusive evidence:
Very often, there are contemporary documents that to
some extent record what happened, and there is sworn oral evidence given by witnesses
in court. There may be conflicts between oral evidence given by different
witnesses, and some or all of this evidence may conflict to a greater or lesser
extent with contemporary records. Opposing accounts of what people did may
appear to conform to how one would expect people to behave in the circumstances
in question, or to deviate to a greater or lesser extent from such
expectations. The behavior of the witnesses in giving their evidence may give
an impression of honesty or dishonesty, or of good or poor memory. Evidentiary
conflicts of these kinds are not, and I say cannot be, resolved by explicit
formal reasoning. The conflicting considerations are inconclusive and of
different kinds, and there is no way in which a judge can overtly apply precise
rules of formal reasoning to them in such a way as to determine a conclusion.
And yet, I contend, reasonable albeit fallible decisions are made. (Hodgson
2012, 38)
Hodgson maintains that
plausible reasoning is an important aspect of human rationality. For example, he
also argues that scientists rely on plausible reasoning when they generate
hypotheses, consider suitable experiment methodology, and provisionally select
unrefuted hypotheses (2012, 39). If Hodgson is right, then it would seem that
plausible reasoning is something that would be problematic for algorithmic
agents. According to him, humans are able to make decisions that involve the
weighing of incommensurables, and this is taken as evidence in support of his
thesis that the decisions of humans do not always proceed algorithmically. He
claims that this leaves space for an undetermined form of agency, where the
reasons do not determine the outcome of a decision.
Hodgson accepts the
possibility that, underlying the level of conscious reasoning, there may be an
unconscious algorithmic process which accounts for decisions that, at the
conscious level, involve plausible reasoning. However, he does not take this to
be likely, as, for evolution to have led to consciousness, Hodgson claims,
consciousness must have had a role. Evolution does appear to have favored
consciousness and so we can assume it does have a role. But what could that
role be?
According to Hodgson,
humans can engage in plausible reasoning and make reasonable decisions. In such
cases, they do this by consciously responding to gestalts (2012, 79–113). He
argues that agents experience sets of circumstances as gestalts, in which the
whole experience is greater than the sum of the parts. This experience of the
circumstances as a whole, rather than merely a series of component parts,
influences agents’ decisions. Hodgson claims such appraisal could provide a
role for consciousness. Its role would be to appraise a gestalt in a way that
is a positive contribution to decision making.
So on this view, the
appraisal of the gestalt contributes to the decision-making process. As the
particular features of any whole set of circumstances that require a decision
are unique, Hodgson argues that no law could dictate a response to such a
gestalt, as laws apply to classes of things. Although one could perhaps respond
algorithmically to component aspects of a set of circumstances that are factors
in a decision, one could not respond algorithmically to the gestalt, to the
experience of the whole situation
(such as that of judge described earlier), due to its uniqueness.
So according to Hodgson,
laws need to apply to classes of things, and so any response to the uniqueness
of the gestalt could not be governed by a law. As humans can indeed respond to
gestalts, Hodgson maintains, it seems unlikely that they are governed by laws
that conclusively dictate action and this makes a deterministic view of agency
less plausible, and makes it more plausible that we are libertarian agents. On
his view, we have the capacity to make decisions that are not wholly determined
by the way we are and the laws of nature.
Hodgson admits he is
making a claim that might be falsified by science, but he argues that science
as it currently stands does not suggest this account is wrong. In doing this,
he points to a scientific worldview – that of quantum physics. This view
suggests the possibility of indeterministic causation in which causes do not
determine their effects but merely raise the probability of a certain effect.
If this interpretation of quantum physics is correct, then determinism is false
and the world is indeterministic. Thus, like Harari, Hodgson accepts that
scientific understanding is of significance for the free will debate, but he
claims his own libertarian view is not refuted by science.
The following seem to be
implications of Hodgson’s work that are particularly salient for a response to
Harari:
1. We
have a capacity for plausible reasoning
2. This
capacity derives from our consciousness
3. We are
indeterministic choosers as we have libertarian free will
The value of
consciousness and free will in a technological dystopia
Will
we be useless in the coming technological dystopia? If we retain some limited
usefulness, will artificially intelligent algorithms find us easy to manage? If
Hodgson is right, then perhaps the answer is “No” to both questions or at least
to the first.
On
Hodgson’s view, our consciousness gives us a capacity that non-conscious AI
cannot attain – we are plausible reasoners. This enables us to perform certain
sorts of jobs that these algorithms cannot do, or at least cannot do as well
(such as the judging and the scientific tasks described earlier). However, the
need for plausible reasoning seems to be widespread in the workforce. Consider
the work of an architect, who must balance aesthetic considerations against
considerations pertaining to the bearing of load in a structure. Or the many
occupations that require a weighing of ethical considerations against
commercial factors.10 If one were to accept Hodgson’s view, perhaps
it is unsurprising that the job of social worker, with its many competing
considerations, has been regarded as one that is at very low risk of automation
(Frey and Osborne 2017). As plausible reasoners, we may be significantly more
useful in the labor markets of the future than Harari envisages, as the need
for plausible reasoning seems to be fairly pervasive in many occupations.
Even
if artificial algorithmic agents surpass us in many intelligence domains, and
ultimately dominate us, so long as they lack consciousness they might find us
useful for tasks involving plausible reasoning (if such tasks enable them to
pursue their aims).11 Assuming the inability to engage in plausible
reasoning does not prevent the rise of the algorithms, but they might need us
for the many kinds of work that require the weighing of incommensurables. To
the extent that plausible reasoning is useful in the formation of beliefs, we
might be used as epistemic workers – helping the artificially intelligent
algorithms to know about the world in ways that they cannot otherwise. To the
extent that these algorithms aim to act so as to accomplish tasks that involve
the weighing of incommensurables, they might need to rely on our advice as to
how they should act, or perhaps delegate that work to us. Rather than our
becoming useless, it seems that the algorithms may need us in a novel division
of labor.
Will
the algorithms successfully manage us? Perhaps the economic value of conscious
plausible reasoning goes some way in explaining why the algorithms might be
motivated to engage in this kind of management – they might need us for certain
kinds of work that are instrumental to their aims, and need to manage us to
ensure that we do it.
In
relation to the management scenario, Harari envisages algorithmic forms of AI
that have knowledge about us that exceeds our self-knowledge.12
According to Harari, because the algorithms know us so well, we will see them
as better decision-makers and will entrust our decision-making to them. Even if
we thought it was prudent13 to trust the algorithms, how far would
we go in entrusting our decisions to them? That is an interesting empirical question
and perhaps we should not assume that people will act in a way that they deem
prudent, and thus defer to algorithms. But from now on, I will assume this
trust.
Harari
claims that the algorithms may use their knowledge of us in order to manipulate
us, but he does not clearly specify the means of manipulation. He appears to
envisage that they will use their knowledge to give us advice that, based on
their intimate knowledge of us, we are likely to follow (rather than use a
manipulative method such as direct brain stimulation) (Harari 2016, 334–46).
This seems to involve the algorithms using their knowledge to predict how we
will react to their advice.
If
Hodgson is right about libertarian free will, there will be a limit to the
extent to which our choices can be predicted, no matter how much is
known about us as agents. Libertarian choices are undetermined, even by the agent themselves, and to
know everything about how an agent is prior to a choice is not to know how they
will choose.
Similarly
to another libertarian, Kane (1998, 74–77), Hodgson envisages the possibility
of undetermined self-forming actions, in which agents make choices that have
the effect of changing their characters (2012, 175). Thus, after a particular
self-forming action, some of what the artificially intelligent manager knew
about the agent prior to the choice might no longer be true of the post-choice
agent. So, perhaps the managing algorithm would now need to wait and get more
information about the post-choice agent before it could be said that they know
that transformed agent. A similar problem would recur after the next
self-forming action. Depending on how frequently self-forming actions take
place, this might be problematic for algorithmic managers, and of course this
issue would only compound any difficulties they had in reassessing the human
agent after unforeseen changes brought about by the agent’s environment.
Perhaps
the algorithms would not need to know with
certainty how humans will act in order to manage them effectively, and
maybe fairly good predictability would suffice for their purposes. Indeed,
today some members of the species Homo
sapiens are able to manage others, despite the issues outlined in this
section. Furthermore, there have been concerns expressed about manipulation by
AI that already occurs (Yearsley 2017). That said, if Hodgson is right, our
libertarian free will would appear to put some limitations on the algorithms’
management capabilities. It is unclear, however, how much of an impediment this
would be.
It
is worth noting that this impediment might not be altogether a good thing for Homo sapiens, as the management issue
might detract from our economic value in the employment market. However, so
long as members of our species are being managed, it seems we are being
regarded as in some way useful.
If
one were to accept Hodgson’s reasoning, and to conclude that there is something
useful about plausible reasoning, this might be a reason to regard Harari’s
third scenario as a little more plausible. If Homo deus is an enhanced version of Homo sapiens, then it will presumably retain the capacity for
plausible reasoning, and possess that capacity together with other enhanced
capacities. This might confer an advantage over non-conscious AI, at least in some
situations. Homo deus might still
value the plausible reasoning of Homo
sapiens, and use members of that subordinate species for work that requires
this capacity.
Hodgson’s
work seems to suggest that there may be something that conscious agents can do
that non-conscious algorithmic AI, in
principle, cannot do. This, of course, is a strong claim. However, even if
one were not convinced that it is, in principle, impossible for a non-conscious
algorithmic AI to achieve something as economically useful as the capacity for
plausible reasoning, one might still see merit in Hodgson’s arguments. One
might take the view that Hodgson’s approach at least identifies issues, such as
the rational resolution of incommensurables, that will be very difficult
hurdles for those engaged in the production of AI. This then has consequences
for Harari’s three scenarios. It might mean that the worst of the economic
effects on Homo sapiens are likely to
be delayed until the hurdles have been cleared.
It
must be acknowledged that the arguments in this paper are predicated on the
assumption that Harari is right in assuming that artificial agents will operate
algorithmically and are not likely to become conscious in a time frame that is
worth considering. So if conscious AI that could engage in plausible reasoning
were to be developed, humans might then become “useless” in the manner
envisaged by Harari. Any reassurance provided from this paper relates only to a
period in which AI is not conscious, and so the argument will be of greater
comfort to a person who thinks that non-conscious AI is impossible, or very
difficult to achieve, than to one who thinks its development is imminent.
A
final point is worth noting prior to concluding. As has been argued, in each of
the scenarios Hodgson’s form of libertarian agency appears to confer an
economic advantage, at least over determined agents of the algorithmic variety.
Whilst economists have turned their mind to the issues of free will (Altman
2006), as far as I am aware, this is the first paper to argue for the economic
benefit to the agent of possessing a form of libertarian free will. It
consequently represents a novel (and worldly) step in debates about free will.
Conclusion
Through
grim speculation on the way that the algorithmic nature of our agency may
negatively impact on our ability to compete with AI, Harari raises the question
of whether consciousness and free will are worth anything. On Harari’s analysis
free will is an illusion, consciousness has little economic value, and our
future looks bleak.
However,
if Hodgson’s arguments about consciousness, plausible reasoning and free will
are sound, then we are not algorithmic choosers. This might not avert some of
the dystopian future scenarios that Harari considers, but the conscious
exercise of libertarian free will, by way of plausible reasoning, might
mitigate some of the negative consequences. In particular, consciousness might
give us a valuable capacity that helps us decide what to believe, and what to
do, when faced with incommensurables. Crucially, this would be a capacity that
non-conscious AI lacks.
Homo sapiens might therefore retain some value in a technological
dystopia because of the continuing value of consciousness and free will.
Hitherto, those engaged in the free will debate have generally addressed other,
perhaps loftier, questions about freedom, but if one now asks what turns on the
issue of whether we have free will, on the view presented here the answer might
be: our livelihoods.
Acknowledgments
I
am grateful to Paul Dower, Neil Levy, David Baker, Jesse Cunningham, Oisin
Deery, Glenn Carruthers, Paul Ham, Mike Bain, Gary Edmond and Meredith
Burgmann, Andrew Haywards, Sam Wyper, Kevin Walton, and Michele Loi for their
useful comments about all or parts of drafts of the paper. I am also grateful
to David Christian and Elle Hrobat from the Big History Institute, at Macquarie
University, for their help and to Russell Blackford and the anonymous reviewers
for their constructive suggestions.
Notes
1.
Harari also publishes on military history.
2.
The word “dogma” has a religious association and this does not seem to be
unintentional. Harari sees the algorithmic view of human agency as part of an
emerging religious worldview which he calls “Dataism” in which the universe is
thought to consist of “data flows, and the value of any phenomenon or entity is
determined by its contribution to data processing” ((Harari 2016, 367). Harari
calls for a questioning of this worldview (Harari 2016, 394).
3.
See Hodgson (2012, 82–94) for an example.
4.
For an edited collection with a further set of responses to Hodgson’s last book
see McCay and Sevel (forthcoming).
5.
See Bostrom (2005) for an overview of this tradition.
6. His
assumption has been challenged by Gray (2016).
7. As noted by Chalmers, functionalist theories of consciousness maintain that “what
matters to consciousness is not biological makeup but causal structure and
causal role, so that a nonbiological system can be conscious as long as it is
organized correctly” (2015, 202). For Harari, we are both algorithms and
conscious. Having accepted that biological algorithmic systems such as humans
can be conscious, perhaps Harari should take more seriously the possibility
that highly sophisticated non-biological algorithmic systems could be organized
in such a way as to become conscious if he does not have a reason to reject the
functionalist view of consciousness.
8. Damasio
(2016) has objected to Harari’s assumption about the insignificance of the
substrate.
9. As
Harari does not envisage conscious AI, presumably he does not envisage
conscious understanding, knowing, or learning by AI devices.
10. I
am grateful to one of the anonymous reviewers for suggesting these examples.
11. As
Bostrom has noted, there are very significant difficulties in predicting the
aims of any future artificial intelligence (2014, 127–39).
12. For
an account of some practical computational problems related to such technology
see First (2017).
13. First (2017) has
persuasively argued that it may not be prudent to put too much trust in
algorithms, bearing in mind that the corporate creators of the algorithms may
well create them with their own interests in mind, rather than the interests of
the agent for whom they “facilitate” choice.
References
Altman, M. 2006. Human agency and free will: Choice
and determinism in economics. International Journal of Social Economics 33(10): 677–97.
Bostrom, N. 2005. A
history of transhumanist thought. Journal of
Evolution and Technology 14(1): 1–25.
Bostrom,
N. 2014. Superintelligence: Paths, dangers, strategies. Oxford: Oxford
University Press.
Capes,
J.A. 2012. Review of Rationality + consciousness = free will.
Notre Dame Philosophical Reviews, July 6.
https://ndpr.nd.edu/news/rationality-consciousness-free-will/ (accessed
September 5, 2018).
Carruthers.
G. 2012. Review:Rationality + consciousness = free willby David Hodgson. Metapsychology 16(43), October 23,
2012.
http://metapsychology.mentalhelp.net/poc/view_doc.php?type=book&id=6670&cn=394
(accessed
September 5, 2018.
Chalmers, D.J.
2015. The singularity: A philosophical analysis. In Science fiction and philosophy: From time travel to
superintelligence, ed. S. Schneider, 171–224. Oxford: Blackwell.
Damasio, A. 2016.
We must not accept an algorithmic account of humanity. Huffington Post, June 28 (updated December 6, 2017).
http://www.huffingtonpost.com/antonio-damasio/algorithmic-human-life_b_10699712.html (accessed
September 5, 2018).
Deery,
O. 2015. Rationality + consciousness = free
will, by David Hodgson. Mind 124(493): 347–51.
de
Sio, F.S., and Vincent, N.A. 2015. Rationality
+ consciousness = free will by David Hodgson. Criminal Law and
Philosophy 9(4): 633–44 .
Earp,
B.D. 2013. Does rationality + consciousness = free will? Journal of
Consciousness Studies 20: 248–53.
First, D. 2017.
Will big data algorithms dismantle the foundations of liberalism? AI & Society, June 6. DOI https://doi.org/10.1007/s00146-017-0733-4.
Frey,
C.B., and Osborne, M.A. 2017. The future of employment: How susceptible are
jobs to computerisation? Technological
Forecasting and Social Change 114: 254–80.
Gray, J. 2016.
Humanity Mk 2: Why the future of humanity will be just as purposeless as the
past. New Statesman, October 13.
https://www.newstatesman.com/culture/books/2016/10/humanity-mk-ii-why-future-humanity-will-be-just-purposeless-past (accessed September
5, 2018).
Harari,
Y.N. 2016. Homo deus: A brief history of tomorrow. London: Harvill
Secker.
Hodgson, D. 2012.
Rationality + consciousness = free will. New York: Oxford University
Press.
Kane, R. 1998. The
significance of free will. Oxford: Oxford University Press.
Levy,
N. 2013. Hodgson, David, Rationality +
consciousness = free will. Australasian
Journal of Philosophy 91(1): 183–92.
Levy, N. 2017. Empirical
perspectives on consciousness and its relationship to free will and moral responsibility.
In The
Routledge companion to free will, eds. K. Timpe, M. Griffith, and N. Levy, 434–44. New York and Oxford:
Routledge.
McCay, A., and
Sevel, M., ed. (forthcoming). Free will
and the law: New perspectives. Oxford:
Routledge.
Nagel,
T. 2012. Mind and cosmos: Why the
materialist neo-Darwinian conception of nature is almost certainly false.
New York: Oxford University Press.
White,
V.A., 2013. David Hodgson, Rationality +
consciousness = free will. Philosophy
in Review 33(2): 126–28.
Yearsley,
L. 2017. We need to talk about the power of
AI to manipulate humans. MIT Technology
Review, June 5.
https://www.technologyreview.com/s/608036/we-need-to-talk-about-the-power-of-ai-to-manipulate-humans/ (accessed September
5, 2018).