Title: The Ethical Status of Artificial Agents With and Without Consciousness EIntentionality 91106
1The Ethical Status of Artificial Agents With
and Without ConsciousnessE-Intentionality
9/11/06
- Steve Torrance
- Middlesex University, UK
- and
- University of Sussex, UK
- stevet_at_sussex.ac.uk
2 - This is an expanded version of a talk given at a
conference of the ETHICBOTS project in Naples,
Oct 17-18, 2006. - See S. Torrance The Ethical Status of
Artificial Agents With and Without
Consciousness (extended abstract), in G.
Tamburrin and E. Datteri (eds) Ethics of Human
Interaction with Robotic, Bionic and AI Systems
Concepts and Policies, Napoli Istituto Italiano
per gli Studi Filosofici, 2006. - See also S. Torrance, Ethics and Consciousness
in Artificial Agents, submitted to Artificial
Intelligence and Society
3What this talk covers
- Artificial Agency (AA)
- Artificial Consciousness (AC)
- Artificial Ethics (AE)
- Artificial Intelligence
- our interaction with them
- and our ethical relation to them.
4Artificial X
- One kind of definition-schema
- ? Creating machines which perform in ways which
require X when humans perform in those ways - (or which justify the attribution of X?)
- Outward performance, versus
- psychological reality within?
5Artificial Consciousness
- Artificial Consciousness (AC)
- ? creating machines which perform in ways which
require consciousness when humans perform in
those ways (?) - Where is the psychological reality of
consciousness in this? - ? functional versus phenomenal
consciousness?
6Shallow and deep AC research
- Shallow AC developing functional replications
of consciousness in artificial agents - Without any claim to inherent psychological
reality - Deep AC developing psychologically real
(phenomenal) consciousness
7Continuum or divide?
- Continuum or divide?
- Is deep AC realizable using current
computationally-based technologies (or does it
require biological replications)? - Thin versus thick phenomenality
- (See S.Torrance Two Concepts of Machine
Phenomenality, (to be submitted, JCS)
8Real versus simulated AC -an ethically
significant boundary?
- Psychologically real versus just simulated
artificial consciousness - This appears to mark an ethically significant
boundary - (perhaps unlike the comparable boundary in AI?)
- Not to deny that debates like the Chinese Room
have aroused strong passions over many years - Working in the area of AC
- (unlike working in AI?)
- puts special ethical responsibilities on
shoulders of researchers
9Techno-ethics
- This takes us into the area of techno-ethics
- Reflection on the ethical responsibilities of
those who are involved in technological R D - (including the technologies of artificial agents
(AI, robotics, MC, etc.)) - Broadly, techno-ethics can be define as
- Reflection on how
- we, as developers and users of technologies,
- ought to use such technologies to best meet
- our existing ethical ends,
- within existing ethical frameworks
- Much of the ethics of artificial agent research
comes under the general techno-ethics umbrella
10From techno-ethics to artificial ethics
- Whats special about the artificial agent
research is that the artificial agents so
produced may count (in various senses) as ethical
agents in their own right - This may involve a revision of our existing
ethical conceptions in various ways - Particularly when we are engaged in research in
(progressively deeper) artificial consciousness - Bearing this in mind, we need to distinguish
between techno-ethics and artificial ethics - (The latter may overlap with the former)
11Towards artificial ethics (AE)
- A key puzzle in AE
- Perhaps ethical reality (or real ethical status)
goes together with psychological reality??
12Shallow and deep AE
- Shallow AE
- Developing ways in which the artificial agents we
produce can conform to, simulate, the ethical
constraints we believe desirable - (Perhaps a sub-field of techno-ethics?)
- Deep AE
- Creating beings with inherent ethical status?
- Rights?
- Responsibilities?
- The boundaries between shallow and deep AE may be
perceived as fuzzy - And may be intrinsically fuzzy
13Proliferation
- A reason for taking this issue seriously
- AA, AC, etc. as potential mass-technologies
- Tendency for successful technologies to
proliferate across the globe - What if AC becomes a widely adopted technology?
- This should raise questions both of a
techno-ethical kind and of a kind specific to AE
14Techno-ethical considerations
BENZ 3-WHEELER 1.7 LITRES (GERMANY, 1885)
- The responsibilities of current researchers in
robotics, etc. can be compared to that of the
founding fathers of automobile design or powered
flight - A certain sort of innocence in relation to the
implications of proliferation that might have
been anticipated
CAR WRECK (USA, 2005)
http//www.car-accidents.com/
15Considerations that seem to transcend the merely
techno-ethical
- There may be deep controversies concerning the
ethical status of life-like artificial agents
(shallow / deep AC agents) - There could be enormous shifts in our ethical
landscape - Our conception of the ethical community, of who
we are, may become hotly contested
16Instrumentality
- Instrumental versus intrinsic stance
- Normally we take our technologies as our tools or
instruments - Instrumental/intrinsic division in relation to
psychological reality of consciousness? - As we progress towards deep AC there could be a
blurring of the boundaries between the two - (already seen in a small way with emerging
caring attitudes of humans towards
people-friendly robots) - This is one illustration of the move from
conventional techno-ethics and artificial ethics
17Artificial Ethics (AE)
- AE could be defined as
- The activity of creating systems which perform in
ways which imply (or confer) the possession of
ethical status when humans perform in those ways.
(?) - The emphasis on performance could be questioned
- ? What is the relation between AE and AC?
- What is ethical (moral) status?
18Two key elements of Xs moral status (in the eyes
of Y)
- Xs being the recipient or target of moral
concern by Y (moral consumption) Y ? X - Xs being the source of moral concern towards Y
(moral production) X ? Y
19Two key elements of Xs moral status
- Xs being the recipient or target of moral
concern by Y (moral consumption) - Xs being the source of moral concern towards Y
(moral production)
(a)
moral agent (natural/artificial)
(b)
moral community (totality of moral agents)
20Two key elements of moral status
- Being the recipient or target of moral concern
(moral consumption) - Being the source of moral concern (moral
production)
(a)
moral agent (natural/artificial)
(b)
moral community (totality of moral agents)
21Ethical status in the absence of consciousness
- Trying to refine our conception on the relation
between AC and AE - What difference does consciousness make to
artificial agency? - In order to shed light on this question we need
to investigate - the putative ethical status of artificial agents
(AAs) when (psychologically real) consciousness
is acknowledged to be ABSENT.
22Our ethical interaction with non-conscious
artificial agents
- ?? Could non-conscious artificial agents
- have genuine moral status
- As moral consumers?
- ? (having moral claims on us)
- (b) As moral producers?
- ? (having moral responsibilities towards us (and
themselves))
23A Strong View of AE
- Psychologically real consciousness is necessary
for AAs to be considered BOTH - as genuine moral consumers
- AND
- (b) as genuine moral producers
- AND there are strong constraints on what counts
as psychologically real consciousness. - ? So, on the strong view, non-conscious AAs
will have no real ethical status
24- One way to weaken the strong view
- by accepting weaker criteria for what counts as
psychologically real consciousness - e.g. by saying Of course you need consciousness
for ethical status, but soon robots, etc. will be
conscious in a psychologically real sense.
25A weaker view
- Psychologically real consciousness is NOT
necessary for an AA to be considered - as a genuine moral producer
- (i.e. as having genuine moral responsibilities)
- But it may be necessary for an AA to be
considered - (b) as a genuine moral consumer
- (i.e. as having genuine moral claims on the moral
community)
26A version of the weaker view
- A version of the weaker view is to be found in
- Floridi, L. and Sanders, J. 2004. On the Morality
of Artificial Agents, Minds and Machines, 14(3)
349-379. - Floridi Sanders Some (quite weak kinds of)
artificial agents may be considered as having a
genuine kind of moral accountability - even if not moral responsibility in
- a full-blooded sense
- (i.e. this kind of moral status may attach to
such agents quite independently of their status
as conscious agents)
27Examining the strong view
- See Steve Torrance, Ethics and Consciousness in
Artificial Agents, - Submitted, Artificial Intelligence and Society
- Being a fully morally responsible agent requires
- empathetic intelligence or rationality
- moral emotions or sensibilities
- ? These seem to require presence of
psychologically real consciousness
BUT.
28Shallow artificial ethics a paradox
- Paradox
- Even if not conscious, we will expect artificial
agents to behave responsibly - ?To perform outwardly to ethical standards of
conduct - This creates an urgent and very challenging
programme of research for now - ? developing appropriate shallow ethical
simulations
29Locus of responsibility
- Where would the locus of responsibility of such
systems lie? - For example, when they break down, give wrong
advice, etc? - On current consensus With designers, operators
rather than with AA itself. - If only with human designers/users, then such
moral AAs dont seem to have genuine moral
status even as moral producers? - BUT
30Moral implications of increasing cognitive
superiority of AAs
- Well communicate with artificial agents (AAs) in
richer and subtler ways - We may look to AAs for moral advice and support
- We may defer to their normative decisions
- E.g when multiplicity of factors require superior
cognitive powers to humans - ? Automated moral pilot systems?
31Non-conscious AAs as moral producers
- None of these properties seem to require
consciousness - So the strong view seems to be in doubt?
- Perhaps non-conscious AAs can be genuine moral
producers - On the question of When can we trust a moral
judgment given by a machine? - ?See Blay Whitby, Computing Machinery and
Morality - submitted, AI and Society
32So
- So non-conscious artificial agents perhaps could
be genuine moral producers - At least in limited sorts of ways
33- Ive challenged this in my paper Ethics and
Consciousness in Artificial Agents - I said that having the capacity for genuine
morally responsible judgment and action require a
kind of empathic rationality -
- And its difficult to see how such empathic
rationality could exist in a being which didnt
have psychologically real consciousness - but Im far from sure
34- In any case, it will be a hard and complex job to
ensure that they simulate moral production in an
ethically acceptable way.
35Non-conscious AAs as moral consumers
- What about non-conscious AAs as moral consumers?
- (i.e. as candidates for our moral concern)?
- Could it ever be rational for us to consider
ourselves as having genuine moral obligations
towards non-conscious AAs?
36Consciousness and moral consumption
- At first sight being a true moral consumer
seems to require being able to consciously
experience pain, distress, need, satisfaction,
joy, sorrow, etc. - i.e. psychologically real consciousness
- Otherwise why waste resources?
- ? The coach crash scenario
BUT
37The case of property ownership
- AAs may come to have interests which we may be
legally (and morally?) obliged to respect - Andrew Martin robot in Bicentennial Man
- Acquires (through courts) legal entitlement to
own property in his own person
38Bicentennial Man
- Household android is acquired by Martin family
christened Andrew - His decorative products
- exquisitely crafted from driftwood
- become highly prized collectors' items
39Bicentennial Man (2)
- Andrews owner wins for him the legal right to
have bank account and legally own the wealth
accumulated from sales of his artworks, (though
still deemed a machine) - Conflict between Andrew and his owner who
refuses to give him his freedom leads to his
moving to his own home
40 - Andrew, arguably, has moral, not just legal
rights to his property - It would be morally wrong for us not to respect
them (e.g. to steal from him) - His rights to maintain his property
- (and our obligation not infringe those rights)
- does not depend on our attributing
consciousness to him
41A case of robot moral (not just legal) rights?
- Andrew, arguably, has moral, not just legal
rights to his property - Would it not be morally wrong for us not to
respect his legal rights? - (morally wrong, e.g., to steal from him?)
42Does it matter if he is non-conscious?
- Arguably, Andrews moral rights to maintain his
property - (and our moral obligation to not infringe those
rights) - do not depend on our attributing
consciousness to him
43 - On the legal status of artificial agents, see
- David Calverley, Imagining a Non-Biological
Machine as a Legal Person, - Submitted, Artificial Intelligence and Society
- For further related discussion of Asimovs
Bicentennial Man, see - Susan Leigh Anderson, Asimovs Three Laws of
Robotics and Machine Metaethics - ibid.
44Conclusions
- We need to distinguish between shallow and deep
AC and AE - We need to distinguish techno-ethics from
artificial ethics (especially strong AE) - There seems to be a link between an artificial
agents status as a conscious being and its
status as an ethical being - A strong view of AC says that genuine ethical
status in artificial agents (both as ethical
consumers and ethical producers) requires
psychologically real consciousness in such agents.
45Conclusions,continued
- 5. Questions can be raised about the strong view
- - (automated ethical advisors property
ownership) - 6. There are many important ways in which a kind
of (shallow) ethics has to be developed for
present day and future non-conscious agents. - 7. But in an ultimate, deep sense, perhaps AC
and AE go together closely - (NB In my paper Ethics and Consciousness in
Artificial Agents - I defend the strong view much more robustly, as
the organic view.)