Could ChatGPT Be a Cognitive Prosthesis?

Source: Anna Shvets / Pexels

The dream—and, for many, the nightmare—of artificial intelligence is the supercharging of our cognitive abilities with digital technology. Your brain on steroids, as some have put it. But what if the chief impact of systems like ChatGPT is not enhancing the already intelligent mind, but instead helping the least cognitively skilled move up toward the level of the average? This scenario, mooted recently by the economist and writer Tim Harford, implies that chatbots could be a kind of prosthetic, but for intelligence rather than sensing or mobility.

Scales of Human "Intelligence"

“Intelligence” is a fraught topic in psychology. There is no agreement as to whether there is such a thing as “general intelligence” in humans (or any other animal). However, it seems pretty clear that there is a distribution of cognitive ability across human populations. Some people have high cognitive skills, some have less, and most are in the middle. (The causes and measures of the distribution are complex, not well understood, and not relevant here.)

Harford, being an economist, framed the idea of chatbots as a cognitive leveler in economic terms. Drawing on the history of technological impacts on job skills, he argues that new tech tends to have two types of effects: It can provide a boost to the already skilled worker—what economists call “skill-biased technological change.” (He cites the spreadsheet as a tool that was deployed rapidly by skilled accountants and which led to large gains for them.) But tech can also empower low-skill workers. Harford argues ChatGPT could prove to follow the latter trajectory.

Chatbots are undeniably good at putting together coherent linguistic expression. While they are hardly perfect, ChatGPT and its ilk can produce plausible responses to questions that many humans would otherwise have no idea how to answer. They are already good enough to be of great potential use to people with lower cognitive abilities. Consider email. In most parts of today's knowledge economy, email is the indispensable currency of daily life. How much would one improve one’s job prospects, not to say one’s social network and self esteem, if one could compose an error-free, persuasive email with the assistance of AI and its impeccable grammar, and perhaps also know when and to whom to send it? Without such aid, an individual might send the wrong message—or not send it at all. I believe marketers will soon realize this and move to package the basic technology of ChatGPT in ways that target this kind of task.

Current Chatbots May Be as Good as It Gets

If chatbots are useful to those with low cognitive abilities, it doesn't necessarily follow that they will also be useful for those with typical or high cognitive ability. As is now well known, ChatGPT and its ilk are prone to "hallucinate" facts and knowledge. The creators of these systems themselves proclaim clear warnings about how unreliable they can be.

Current AI chatbots may also be as good as it gets. Other highly-touted digital technologies such as virtual reality, cryptocurrency, NFTs and quantum computing have mostly failed to live up to the initial, world-changing hype. I am personally skeptical that AI based on deep learning will improve much. As I have argued here previously, the “guardrails” and fact-checking the systems will require to reach true reliability will defeat the whole purpose of machine learning, which is based on the idea that machines can gain essential knowledge about the world solely through massive ingestion of past events. Ultimately, I believe the efficiency of machine learning will be mostly offset by the need to correct for errors one by one, along with the need for constant retraining in light of new events. Without major new breakthroughs in the basic technology, those of average and higher cognitive ability may have little use for them in most settings. But the systems can already serve as the basis for a cognitive prosthesis.

Cognitive Prosthetics

I want to be clear: I am not suggesting that a chatbot has the same intelligence as an average person. Indeed, in almost every way, a human with low or even very low cognitive ability far outmatches any AI system. Merely having a body in which the brain is housed affords us diverse forms of knowledge—smells, sounds, motor interactions, social interactions—that put all of us in an entirely different league of intelligence compared to any machine. I also not suggesting that we should give up on demanding more reliable AI.

What I am suggesting is that chatbots can help enable typical—rather than exceptional—cognitive skills that are suited for daily life. This is what a prosthesis does: It approximates normal function. A chatbot can provide elements of knowledge, especially written language, that can help those with lower cognitive skills get along in today’s world.

Prostheses are a good metaphor for chatbots because neither is useful in all settings; a prosthetic limb needs to be removed in the shower, for example. So too with cognitive enhancement: AI advantages are harder to deploy in person, though I believe clever adaptations will emerge. In addition to email, chatbots could help low-performing students at the high school and college level avoid failure and dropping out—if society allows it.

Artificial Intelligence Essential Reads

How AI Bias Impacts Medical Diagnosis

Can We Feel Empathy for AI?

Possible Futures

The potential impact of ChatGPT on the human cognitive landscape could be profound. Sellers of these systems could become predatory; those least able to understand the pros and cons of a chatbot could be manipulated into buying overpriced systems. Indeed, as Harford points out, tech advantages for low skilled workers have in the past been coopted by capitalists. Strong regulations are needed to prevent this and ensure that anyone seeking a cognitive prosthesis does so solely of their own accord.

The psychological and social consequences could be just as momentous as the economic ones. If chatbot technology is sufficiently integrated into the self somehow, does it become part of one’s identity, like a prosthetic limb? Is it enough for someone to make a query that is answered by an integrated bot for one to claim the knowledge produced as one’s own? Is it legitimate for someone to be admitted to college with the support of such a prosthesis?

We may be surprised about where AI takes us. Not to galaxy brains for those at the higher end of the distribution, but instead to a narrowing of the distribution, and everything that might entail.

Copyright © 2023 Daniel Graham. Unauthorized reproduction of any content on this page is forbidden. For reprint requests, email reprints@internetinyourhead.com.