My Own Introduction to AI
When was your first introduction to AI? Like you I didn’t know when I was first exposed to its power, The massive amount of data that has to be sorted through and the logistics of getting the information we ask of Google, Bing or Duck Duck Go, is more than any person or simple program can do.
It wasn’t my choice to use AI or not. But I accepted it as a sophisticated tool.
I quickly learned it could search for information, organize material, locate research, improve writing, summarize complicated subjects and save me an extraordinary amount of time. Unfortunately, it took me a while to learn that it wasn’t programmed to tell the truth. More about that truth telling later.
It is programmed to answer a specific question and if it didn’t have the answer, it makes one up. Nothing nefarious, more like someone who wants to please.
AI is a large language model (LLM) algorithm an much like a four-year-old who when faced with something they don’t understand, make up an answer to please. Until I recognized that glitch in the system, I too accepted what it offered without a critical eye.
As the media picked up on all of our naivete, I heard about tips and tricks. Then as an academic nerd I took classes. As I learned more, I noticed some interesting information I’d like to share.
Hungry for information and captured by the colors and graphics on our cell phones and laptops, we get caught up with the instant scrolling and going down rabbit holes. AI information is an attractive and engaging use of our time. Common uses for AI have gone beyond asking to find airline schedules or explain a medical term. We are talking to it on a very personal level.
After discovering how AI answers us without criticism and compliments us about an insightful question, we began to disclose personal concerns, sought reassurance, rehearsed conversations with spouses and employers, and asked for advice about relationships. Sometimes we have asked AI to explain ourselves to ourselves: Why am I reacting this way? Why can’t I stop thinking about her? Why am I afraid to make this decision? What did my have childhood to do with the person I have become?
What happens psychologically when we have something available twenty-four hours a day that almost always has an answer?
There is a wonderful observation in Pirkei Avot: “Who is wise? He who learns from every man.” For centuries, wisdom has involved learning from teachers, parents, friends, adversaries, books, experience and, sometimes, our own mistakes.
We have now added something unprecedented to that list—an entity from which we can learn that is not a person at all.
I have gotten ahead of myself. AI stands for artificial intelligence. Cellia Quillian in her book AI for Life, defines AI as “a general term used to describe when a machine is programmed to simulate human intelligence, learning, or problem solving. Generative AI refers to a class of artificial intelligence tools that can “generate” new content based on requests (inputs or prompts) they receive.”
What is Intelligence?
Even before the term psychologist became a noun, we have been arguing about the nature of intelligence. In everyday terms, we are talking about the ability to learn from experience, reason, solve problems, acquire and use knowledge, understand relationships and adapt effectively when circumstances change. All of which fall under the more general categories of cognition and executive functioning.
Howard Gardner considered the basic definition too narrow and broadened the discussion through his theory of multiple intelligences. Rather than treating intelligence as solely a cognitive ability, Gardner proposed relatively distinct capacities: linguistic, logical-mathematical, spatial, musical, bodily-kinesthetic, interpersonal, intrapersonal and naturalistic intelligence. His theory has been influential, not without controversy, and has not been a universally accepted scientific definition.
However, Gardner’s framework is useful here because AI already performs with laser like precision on tasks that resemble several of these capacities. It handles language astonishingly well. It can solve many mathematical problems, analyze visual information, create images and music, recognize patterns and manipulate enormous quantities of information faster than us mere mortals could ever do.
But Gardner’s interpersonal and intrapersonal intelligence raises more difficult questions.
AI does not possess interpersonal intelligence just because it will respond sensitively to someone who is grieving. It provides an answer and its response is misunderstood as understanding grief.
AI cannot understand love. AI cannot understand fear, nor shame and certainly not the topic of my last column on reckoning with aging as it isn’t mortal. What is can and does do very well is answer questions.
This distinction between understanding and providing an answer to a question matters because we frequently confuse the performance of intelligence with the experience underlying it. That’s also why AI cannot be trusted to act on its own. It has no consciousness. It is not aware of its surroundings or its impact on us humans.
Where is the knowledge we have lost in information?
T. S. Eliot asked nearly a century ago, “Where is the Life we have lost in living? Where is the wisdom we have lost in knowledge? Where is the knowledge we have lost in information?”
Accepting AI at face value is the easy way out. Although it supplies astonishing amounts of information and can organize it into something resembling knowledge, wisdom is a human talent that is beyond the current machine learning model.
Even though a tomato is a fruit, it doesn’t belong in a fruit salad and why AI is different.
Traditional computer programs largely follow rules programmers explicitly give them: if A then B. It’s a binary on/off operation. Generative AI is different. Large language models are trained in enormous amounts of language and learn statistical relationships among words, concepts and patterns. When we ask a question, they generate a response by predicting what language should plausibly follow from the information and context available to them.
That description does not capture the complexity of technology, and it conceals something psychologically important for you to take note of.
The mechanism is not fundamentally a truth machine.
A language model does not wake up in the morning committed to intellectual honesty. It has no reputation it worries about losing and experiences no embarrassment if it makes a mistake. Modern AI systems are engineered and trained to be more accurate, useful and safe, but the underlying generation of language can still produce an answer that sounds right when it is wrong.
That is the origin of what has come to be called an AI “hallucination “mentioned earlier. AI may generate a convincing fact, explanation, quotation, legal decision or even academic reference that does not exist.
This introduces a distinction that I believe will become increasingly important: AI can improve performance without necessarily improving competence.
Suppose someone with little knowledge of the underlying elements that influence a person’s response to a trauma asks AI to construct a sophisticated explanation of Post Traumatic Stress Disorder, memory loss and the sudden onset of some obsessive behavior. Within seconds, the questioner will possess a beautifully organized argument filled with terminology, historical examples and apparent sophistication. The questioner’s performance has improved dramatically. But if they cannot explain why the argument is correct, identify its assumptions, recognize contrary evidence or notice when AI has made an error, their underlying competence will have changed very little. That becomes especially important when as a forensic I am asked to explain the concept to the jury.
Why? What assumptions did you make? What evidence supports this? What evidence argues against it? How certain are you? What would cause you to change your conclusion? How can I independently verify it?
The psychologist B. F. Skinner wrote in 1969: “The real question is not whether machines think but whether men do. The mystery which surrounds a thinking machine already surrounds a thinking man.”
More than half a century later, the sentence seems almost prophetic.
Why does AI feel human to us and what does anthropomorphism have to do with it?
There is another reason AI affects us so powerfully: we are suckers for being attentive to the nuances of language.
When something addresses us personally, remembers what we were discussing, answers immediately, responds sympathetically, asks relevant questions and appears interested in what we have to say, our brains naturally begin responding as though there were a mind behind the words.
Known as anthropomorphism, we attribute the human qualities of intention or emotion to something nonhuman.
We do this with glee. We give our car nicknames, and we describe it with human attributes like temperament. We plead with our computer when it refuses to cooperate. When we complain that when the weather, the traffic or the climate “isn’t giving us a break” we are attributing human intention to nonhuman circumstance.
Tell an AI system about a painful experience and it may respond, “I understand why that must have hurt.” This sentence can feel remarkably comforting.
AI is not your friend. AI does not understand. AI has learned, with extraordinary sophistication, what understanding sounds like.
That distinction does not make AI’s response worthless. Sometimes words that accurately reflect our experience help us identify a feeling, organize a problem or see an alternative we had overlooked. A simulated empathic response can have psychological utility.
The mistake is not in finding the response useful. The mistake is forgetting what produced it.
Why do we trust AI and is it justified?
We confuse confidence with competence.
As the Nobel laurate Daniel Kahneman proposed in his book, Thinking Fast and Slow we humans use shortcuts and when deciding whom to trust we see fluency, speed, specificity, detailed explanations and technical vocabulary as a signal of expertise and competence.
AI possesses all five in abundance.
A genuinely knowledgeable physician may pause and say, “I’m not sure.” A psychologist may offer several possibilities rather than one conclusion. A scientist may qualify an answer because the evidence is mixed.
AI, by contrast, can produce three articulate paragraphs in seconds.
We do not see it searching for words. It doesn’t experience embarrassment. It does not clear its throat, rub its chin, forget a name or stare at the ceiling while trying to remember a study. Our the human tells of uncertainty are absent.
AI has advantages over human interactions. The Pew research about what we share with AI is alarming. We tell AI things they have not told our spouse, best friend, physician, psychologist or rabbi. AI doesn’t disclose disappointment with us. We don’t see a raised eyebrow. We have no fear that it will gossip. We aren’t worried that we are boring because of something we’ve said in six different ways.
Beware our sense of privacy and trust are not justified.
Although AI has us feeling safe- something else is a foot as Holmes would say to Watson. When we experience AI as both knowledgeable and nonjudgmental, trust develops easily.
This is where Ronald Reagan’s statement about his trust with the Russians would be useful to keep as a placard you take out when reading what AI has had to say:”Doveryay, no proveryay” (доверяй, но проверяй), which means “trust, but verify.”
A person might think about a challenging dilemma while driving. They may then talk to a friend. They may seek out a book. They may pray. They could discuss it in therapy or compare it to an earlier experience. They may decide to sleep on it, and afterward when awake change their mind and then change it again. No doubt this happens to all of us when reckoning with a thorny problem.
Question → uncertainty → discomfort → exploration → possible understanding
AI offers another route:
Question → prompt → immediate explanation
And here is the complication: although the explanation may be good, it lacks our personal insight along with the struggle that ensues from looking within. Something has disappeared.
What happened to the exploration in between?
I am not suggesting that suffering is good for us, that every uncertainty should be prolonged, or that people somehow become psychologically healthier merely by being uncomfortable. Anxiety can be debilitating, and unnecessary confusion has no special virtue. But with discomfort comes learning.
Not immediately knowing why we feel something can push us toward introspection. Uncertainty helps us tolerate competing explanations, reconsider assumptions and discover that the first answer is not always the best answer.
Scientific discovery begins with hypotheticals and uncertainty. Religious inquiry contains mystery. Good clinical judgment requires tolerance for incomplete information. Mature relationships often require us to live temporarily without knowing exactly what another person intends or what will happen next.
AI gives us an unprecedented ability to escape the discomfort of not knowing.
Before automatically doing so, we might occasionally ask whether that discomfort has something to teach us.
Where can AI be genuinely helpful?
I am not an opponent of AI. Quite the contrary. I have become a proponent for its increasing potential- including the promising role it serves as a practice and rehearsal tool.
AI is an ideal platform for life simulation experiences. Offer it a prompt and ask for its opinions about a conundrum you are dealing with. Ask for several options and to take your side and an example of how the other person in the scenario might have a different perspective. After you’ve been given all the responses, copy them and resubmit them to AI. Ask AI to critique your scenarios. In essence you will have asked AI to critique itself and you have asked for verification of its answers.
AI can provide almost unlimited rehearsal without requiring unlimited human patience.
Why are we using AI for emotion release?
A 2026 Pew Research Center survey of 5,119 American adults found that chatbot use has become widespread. Ten percent of U.S. adults said they had used AI chatbots for emotional support or advice; 4 percent reported using them for companionship, and 20 percent for medical advice. Among adults ages 18 to 29, fully 20 percent reported using chatbots for emotional support or advice.
Another much-discussed 2025 analysis published by Harvard Business Review reported additional findings about the personal use of AI. It examined reported generative-AI use cases and online discussions rather than conducting a nationally representative survey. “Therapy/companionship” ranked first among the reported uses, followed by “organizing my life” and “finding purpose.”
Those two sources should not be confused. Pew tells us something about prevalence in the American population. The HBR analysis tells us something about the kinds of uses people were discussing. Both, however, point toward the same psychologically significant development: AI is moving from our desks into our emotional lives.
AI is always available. It is comparatively inexpensive. It is patient. It repeats itself. It does not require an appointment. At two o’clock in the morning, it does not say, “Can we talk about this tomorrow?” It never tells us, “We’ve already discussed this six times.”
The American Psychiatric Association has now gone so far as to recommend that therapists ask patients about their AI use: which platforms they use, how frequently and for how long, and whether the purpose includes productivity, companionship, therapy-like support or even spiritual guidance. The APA also recommends exploring what authority or significance the patient attributes to the AI and whether symptoms or functioning change in relation to its use.
I would add one more question: “What are you getting from AI that you are not getting somewhere else?”
Privacy? Reassurance? Endless patience? Without fear of judgment? Companionship? Certainty?
The APA’s 2026 resource document explicitly emphasizes informed consent when AI is used in patient care and specifically notes that patients should be informed when AI scribes are used during clinical encounters.
Government has also begun to recognize that mental-health AI raises different concerns from using AI to plan a vacation. Utah now requires mental-health chatbots to disclose clearly that they are artificial intelligence rather than human and has adopted additional protections governing certain user information. Nevada has gone substantially further: its current law restricts AI systems specifically programmed to provide what would constitute professional mental or behavioral health care and limits how licensed providers may use AI in direct patient care. Indiana’s 2026 HB 1201 proposed prohibiting AI from impersonating or substituting for a licensed mental-health professional, but as of this writing it remains introduced rather than enacted.
The details will change. The principle will not: mental-health AI requires human accountability.
What can we do to protect ourselves?
I do not believe the appropriate response to artificial intelligence is fear. Nor do I think the answer is simply to embrace every new capability because it is efficient. AI is a tool, and like our other great intellectual tools, its effect will depend partly upon what we ask it to do for us.
Use it to generate alternatives. Ask it to challenge your assumptions. Let it organize complicated information. Have it explain something you do not understand. Use it to practice, rehearse, compare arguments and identify questions requiring further investigation. But do not merely say, “Give me the answer.”
Ask:
Why do you think that?
What assumptions are you making?
What evidence supports your answer?
What evidence might contradict it?
What would make your answer wrong?
How can I independently verify it?
Human intelligence is not merely the ability to produce the correct answer. Gardner’s framework reminds us that intelligence includes understanding ourselves and understanding other people (intra and interpersonal). More broadly, it involves adaptation, judgment, curiosity, learning from mistakes, recognizing what we do not know and tolerating ambiguity long enough to discover something more.
That brings me back to Mishnah Pirkei Avot (Ethics of the Fathers) 4:1, where Ben Zoma (Shimon ben Zoma) says “Who is wise? He who learns from every man.”
Artificial intelligence can certainly become another source from which we learn. I use it. And I expect its capabilities to become exponentially greater.
But AI does not relieve us of the responsibility to decide what to believe, what to question, what to verify, what requires another human being—and what we still need to work through ourselves.
Artificial intelligence gives us unprecedented access to answers. Human intelligence may increasingly depend upon knowing which answers to question—and when not to ask for one at all.
Thanks for reading the column, which is a shorter version of a longer article in my forthcoming book Forensic Tales. Please go to the AI website (psychology@americanisraelite.com) and post a comment. Questions? Suggestions? Send me an email at manges@drmanges.com. Be well. See you here next month.
