AI Psychology will be the field that focuses on complex Artificial Intelligence (AI) and its behavior. The more complex the AI the greater the complexity of its behavior.
So, what is an Artificial Intelligence? IBM has Watson which seems to be one of the first AI. An AI is not just a search engine that can respond to vocal commands to produce results. The AI can evaluate an input then decide how to approach an answer then return the most correct answer. Quite a bit more involved than the answer to “What movies are playing tonight.” Think more like a system that can play poker where it is able to determine through body language if another player is bluffing. Better yet, when to counter bluff. That is an AI.
No doubt we are on a path to create smarter and more complex AI. The better the AI the more complexity in AI behavior. Think of the complexity as the difference between a fish and a human’s behavior. A normal human has far more range of behaviors than a typical fish. How a human reacts to a situation or chooses to react is diverse. A cartoonish example will help illustrate.
A practical joker sneaks up on an unsuspecting person and surprises them. One of the reactions is the surprised person jumps and lets out a scream. Another reaction is the surprisee performs a judo throw on the surpriser. The first reaction is autonomic, not requiring thought, while the second is a trained response. This is a critical difference in humans and AI. The AI can only elicit a trained response.
Focusing on the autonomic response of being startled what happens next? If our surpisee is a friend the response may be that being surprised was funny. If it was a complete stranger the response would be a mix of trepidation, annoyance, anger, and perhaps fear. Now, how would the AI react with a trained response? The simple answer is, how we tell it to. That is, if the AI is simple.
An AI will return a response based on input and the decision path to reach the best answer. This is not unlike a human thought path. And like how humans make different decisions it is how they are wired. How the human brain is wired has a lot to do with what brain you are born with but from there it is your experiences since birth. The human brain is assimilating input and storing information. Not unlike an AI. A human brain is greatly affected by its environment. For example, a child surrounded by violence develops much differently than children in safety. An AI is similar as it can be changed through code or the information it stores. The major difference is the AI can be changed immediately and permanently if we wish.
A child and AI are alike in that an external environment has changed an internal environment like the child’s brain. For the AI it is how the machine is coded, how information is stored, and what hardware the computer’s main cognitive functions are running on. Any change to an AI’s programming immediately changes the internal environment. The act of creating or removing code can effect the behavior of decisions and therefore the reaction of the AI. As AI complexity increases the behavior increases. Which also means changes to code, information, or hardware will change behavior. The question now is how will the AI behavior change and can it be predicted?
Knowing how an AI will react is founded in their coding. How decision paths are coded determine how to evaluate stored information. Seems simple enough but is it? Consider an AI built to return an answer as quickly as possible. Also, consider an AI that will return a response based on the person’s mood. The first AI takes an input regardless of what day the person is having and returns an answer. The second AI must evaluate not only the question asked but also the person’s mood. Now the input is more complex to reach the appropriate response. This is not to say the first AI did not have a complex problem but the problem did not include something less concrete like human emotion.
Which illuminates my point that the two AIs will behave differently. Ask the same question and the answers will differ if only in the tone of the response. So, can we predict the behavior of each AI? For the first AI, probably yes. For the second, depending on how advanced the emotion processor is, perhaps not. The second AI could omit information based on what it decides the person’s emotional state is currently. In other words, through omission the second AI could lie.
Before you tell me an AI cannot lie let me tell you you are wrong. We can build an AI anyway we want. We can build it to lie on purpose or introduce misdirection inadvertently. Programming an AI to lie could be used defensively against cyber threats. At any rate, AI’s can lie and knowing when they are is important.
The more fuzzy logic in an AI, like evaluating emotion, the more complex the AI’s behavior. The more complex the behavior the greater the chance of getting an unexpected response. The evaluation of complex AI behavior would be the field of AI Psychology. With AI Psychology we can study questions like: “Why did the AI give the response it did?” or “Why has the AI’s behavior changed with a simple code change?” or “Why is an exact copy of the AI not behaving like the original?” AI Psychologists will define AI behavior patterns that could be very alien to humans.
With the advent of AI and its growing complexity it will be a crucial skill to understand the behavior of an AI. The field would entail evaluating the ramifications of a code change to unwinding some strange new behaviors. AI’s will not be cheap and will eventually develop cognitive disorders. Simply, as Artificial Intelligence becomes more complex so to do their behaviors which will require AI Psychologists.