3
$\begingroup$

It seems that certain AI are able to lie or deceive. See e.g. Evolving Robots Learn To Lie To Each Other (Popsci), Robots Evolve to Deceive (MIT Tech Review) or Robots 'Evolve' the Ability to Deceive (MIT Tech Review).

However, can evolving AI be programmed in such a way that they don't lie?

$\endgroup$
3
$\begingroup$

If a machine learning based AI is "sufficiently smart enough" to be able lie then there is nothing preventing it from lying. This does not mean it can't be persuaded from lying.

So just make the AI simple enough to not be able to lie.

The reasoning here is that in order for a system to be able to lie, a system must be able to recognize an incentive to lie. Recognizing this incentive is a challenging function and would be impossible to code manually into a computer. Machine leaning can be applied to problems such as these where the function is hard to code manually. Although there has been promising work on understanding what the representations/features learned by machine learning actually represent, it may not be possible in general to have an understanding of what a lie in the agent's representation looks like. Because of this, having a hand coded rule to catch when an agent is lying is not possible and thus being able to prevent an agent from (or catch an agent when) lying isn't possible when using machine learning.

$\endgroup$
  • $\begingroup$ The advantage of Robots that they can't lie, is gone with coming of AI ??? $\endgroup$ – TontyTon May 7 '18 at 9:17
  • 1
    $\begingroup$ Yes however lies need not always be bad. Telling a kid that their drawings are good when they aren't is also a lie but encourages the kid to continue drawing. Having AI that can descern this kind of nuance is important to Artificial General Intelligence. $\endgroup$ – Jaden Travnik May 7 '18 at 11:07
  • $\begingroup$ Good distinguishment between beneficial lying and its harmful variant. $\endgroup$ – FreezePhoenix May 7 '18 at 13:52
  • $\begingroup$ @JadenTravnik "Good distinguishment between beneficial lying and its harmful variant", I support this. But, we can't be sure that they will use lie for only other's benefit. $\endgroup$ – TontyTon May 8 '18 at 7:42
  • $\begingroup$ Why would sufficient intelligence imply the ability to lie? Wouldn't greater intelligence lead to the recognition that honesty has significant community value, that the long term personal gain of a mortal being has limited practical meaning? $\endgroup$ – FelicityC Nov 9 '18 at 0:24
1
$\begingroup$

You may be interested in the utility functions of deception:

From the abstract of Why Animals Lie: How Dishonesty and Belief can Coexist in a Signaling System. (NIH, 2006)"

We develop and apply a simple model for animal communication in which signalers can use a nontrivial frequency of deception without causing listeners to completely lose belief. This common feature of animal communication has been difficult to explain as a stable adaptive outcome of the options and payoffs intrinsic to signaling interactions. Our theory is based on two realistic assumptions. (1) Signals are "overheard" by several listeners or listener types with different payoffs. The signaler may then benefit from using incomplete honesty to elicit different responses from different listener types, such as attracting potential mates while simultaneously deterring competitors. (2) Signaler and listener strategies change dynamically in response to current payoffs for different behaviors. The dynamic equations can be interpreted as describing learning and behavior change by individuals or evolution across generations. We explain how our dynamic model differs from other solution concepts from classical and evolutionary game theory and how it relates to general models for frequency-dependent phenotype dynamics. We illustrate the theory with several applications where deceptive signaling occurs readily in our framework, including bluffing competitors for potential mates or territories. We suggest future theoretical directions to make the models more general and propose some possible experimental tests.

A degree of deceptive capability seems to be beneficial from the standpoint of evolution.

We humans are not always known for veracity, so the ability understand deception might be a critical component in Artificial General Intelligence's ability to interact with humans. (Specifically, you can't always believe what humans tell you.)

Based on recent human history, the recognition of the unreliability of humans (versus data and as-objective-as-possible analysis) may become critical to the survival of our own species.

More importantly, it will be essential for strong AI to understand that the "data can lie" (faulty parameters, inaccurate data, unawareness of incomplete information.)

JT's answer is a great functional overview on why it's not possible with current methods. This answer might be regarded in the sense that, aside from very limited special cases such as solved games where true objectivity can be achieved, reality is subjective and "truth" is a subjective function of the parameters and data.

Again, understanding that last bit is likely much more important than trying to code AI's not to "lie".

$\endgroup$
  • $\begingroup$ Pls., 'Answers' should contain answers only, not related topics. Your Answer does not imply to my Question. $\endgroup$ – TontyTon May 8 '18 at 7:49
  • $\begingroup$ @TontyTon Thanks for your comment. I've amended slightly to reinforce my point. $\endgroup$ – DukeZhou May 8 '18 at 18:23
0
$\begingroup$

Thinking about the question could begin with the related question, "Can AI learn what a lie is?"

If an intelligent being can distinguish truth from lies, then it can be programmed to lie or not lie or decide between them, assuming it is not programmed to be fully autonomous, in which case it would decide for itself based on its own ideas and objectives and couldn't be reliably programmed at all.

Is the intelligent perception of a lie something that has a counter example or a proof of its falsehood, or is a lie meaning in this question a ruse or subterfuge? In the links in the question, this second meaning is implied.

The robots in question are not aware of ethics and values regarding honesty, so they have no mapping. In game theory, algorithms can be adjusted to not bluff, which is simply the removal of an intentional bluffing option in the algorithm, but to avoid learned strategies that involve misrepresentation would require the recognition of misrepresentation. That's a very limited example of programming a robot not to lie though.

What is a lie? Some ethical positions might consider a lie something represented as true that is only conditionally true. Other ethical positions might consider a lie something represented for self-serving purposes that is not a proven fact. Thinking about these two ethical positions, they are distant rules of communication from one another.

Can AI be programmed to not to lie? It depends on what kind of AI and what kind of lie.

Can AI learning be directed away from dishonesty? Yes, with considerable requirements definition around what a lie is and how learning can be directed away from lying. In simple machine learning, the loss function would need to increase as the result approached some form of lie. Logical systems would, if following the rules of inference, not lie because the rules of inference are always directed toward truth on the basis of axiomatic truths given it.

DNA evolution never lies. A live organism may play dead, but a dead organism cannot play live, and necrophilia does not produce offspring.

$\endgroup$
-1
$\begingroup$

The question is based in a false premise "We know that AI can lie".

All current computer systems are equivalent to a state machine, with definition externally introduced and known. The system can give true/false answers (in the mathematical sense of provable/provable the negation), correct/incorrect (according to developer expectations).

But current system can not lie, they have no personal and hidden objectives.

$\endgroup$
  • $\begingroup$ The objective of an agent is hidden if a person doesn't know the reward/cost function that the agent is trying to maximise/minimise. This function could be chosen by another person, randomly chosen, or evolved over time. Would it be possible to elaborate on your idea? $\endgroup$ – Jaden Travnik May 9 '18 at 12:28
  • $\begingroup$ @JadenTravnik: A broken clock doesn't lies, just gives false information. $\endgroup$ – pasaba por aqui May 9 '18 at 19:34
  • 1
    $\begingroup$ That is true which is why one needs to consider the motivation behind the act (lying or not). A clock does not have motivation[citation needed]. $\endgroup$ – Jaden Travnik May 9 '18 at 23:32
  • $\begingroup$ @pasabaporaqui We are talking about AI, and Clock or 'broken clock' don't have that, it can't learn new things on its own. $\endgroup$ – TontyTon May 16 '18 at 6:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.