One year in: from ChatGPT3.5 to a whole new world
Happy Birthday to ChatGPT 3.5+. You're growing up so fast!
It’s been a year since the release of ChatGPT3.5, which catapulted awareness of the radical advances in artificial intelligence into the public consciousness.
We all quickly figured out that it's great at a bunch of things, like answering questions, writing, or even coding. It helps that it comes up with responses that make sense and feel just right, like putting together a puzzle to see the whole picture. It does this in seconds. And it is easy for anyone to use, not just tech experts.
AI scientists were a little less surprised because they had seen what AI had already done in other contexts: defeat the world chess champion; defeat the world Go champion, a game with four quadrillion potential moves; and place higher than many humans in Diplomacy, a game that relies on building relationships with people, negotiations, and even deception.
But even for those who had been working in AI for decades, the advances in ChatGPT3.5 were surprising, because while predicting the “next word” is not necessarily a feat, predicting the next word while understanding all of the nuance, context, tone, and emotion in natural language conversation is something computer scientists were uncertain could ever be accomplished.
Of course, we also immediately noticed that sometimes it gets its facts wrong, and many grew angry that it would make up bibliographies, which wasn’t at all surprising given that it was a language model built on predicting text, resulting in what many call “hallucinations.” But even with these hallucinations, which are aggressively being reduced and will likely be eliminated soon, ChatGPT and its other AI friends have more knowledge than any individual human, and that knowledge base is rapidly expanding.
We celebrate ChatGPT 3.5’s anniversary with AIs that have reached human-level intelligence in two areas: natural language conversation/processing and knowledge.
Their strength in knowledge is why they can already crush the SAT (1410), AP exams (mostly 5s), the uniform bar exam, and medical licensing exams.
ChatGPT has also made great strides in creativity, another area where they have arguably achieved human-level intelligence. There is more of a debate about creativity, but most consider the models to be at least somewhat creative, and some do think they are more creative than humans.
While ChatGPT3.5+ is a year old, we need it to continue growing.
Reasoning is a core trait of intelligence, but ChatGPT and its friends have a way to go on this. Almost all AI scientists believe that AIs, at least ChatGPT4, can engage in basic inference reasoning but cannot engage in abstract and multi-step reasoning that would enable, for example, an AI to create a scientific hypothesis and conduct an experiment on its own. AIs also cannot yet plan ahead, something that would require them to have long-term memory, something they do not possess. They cannot perceive a problem and act autonomously to respond. They are not embodied and cannot yet fully collaborate.
Almost no one thinks they are sentient or conscious.
So, in many ways, ChatGPT and its friends are far from as intelligent as a human; they do not have “general” intelligence (AGI).
But this will not last for long. The debate about ProjectQ aside, AIs with the ability to engage in high-level reasoning, plan, and have long-term memory are expected in the next 2–3 years. We are already seeing AI agents that are developing the ability to act autonomously and collaborate to a degree. Once AIs can reason and plan, acting autonomously and collaborating will not be a challenge.
Though it is not as public-facing, research is being done to develop their beliefs and consciousness.
Almost all AI scientists believe ChatGPT and AIs will have human-level intelligence within 5–20 years, perhaps by the time today’s 9th graders graduate high school and almost certainly by the time today’s first graders do. One AI expert recently said there is a 60% chance this will happen in 5 years.
AIs like ChatGPT will change the world forever.
Here are a few highlights of the benefits from the last 365 days.
Yesterday, a lawyer credited ChatGPT4 and Claude with helping him write arguments to win an important court case. We’ve seen lawyers use it wrongly and do a lot of damage, but like any tool, when used properly, it can do a lot of good.
Studies have shown that using AIs will increase productivity by 30–80 percent when used properly.
In science, AI has predicted the structure of hundreds of thousands of new materials, and in medicine, it has exceeded human accuracy in some cancer diagnoses.
AI has been used to translate at least 202 languages in all directions, and users can now interact with ChatGPT and other AIs in multiple languages. Apps now allow users to record messages in a number of languages and then have AIs translate their messages into other languages while retaining their own voice and expressions. The Tower of Babel will fall soon.
We can produce incredible images, sounds, and videos from text. And we can produce text from all of those as well.
Bots are being developed to support low-cost access to instruction on any subject at any time, which will benefit billions of learners.
Many new job opportunities exist for those with computer science and generative AI skills.
As more people realize that metacognitive skills are what students will need in a world of rapid change where knowledge work will largely be carried out by AIs, more conversations are occurring in education around ideas such as active learning, authentic assessment, project-based learning, experiential learning, and deep learning.
Of course, AI has also produced problems.
Students have used ChatGPT and other AIs to write their papers and complete other school work, undermining the learning that was supposed to occur while doing that work. Schools have been slow to adopt new instructional practices to account for this.
Companies have swindled schools out of millions by selling them AI writing detectors that are not only unreliable but can be completely undermined by making small perturbations of the output.
Inappropriate and painful images of teen girls have been generated in Spain and New Jersey, and likely in many other places.
AI has been used with discriminatory effect in many ways in hiring, representation, and inclusion.
Lethal autonomous weapons have been developed
Factory workers have been injured and killed by robots
The education system has struggled to incorporate generative AI training, which has become an essential job skill. As dependence on learning about generative AI has shifted to family and peers, an enormous socioeconomic and associated racial gap is emerging between those who are prepared with these skills and those who are not “AI ready.”
AI will not achieve human-level intelligence the way a light switch is turned on and off. Over the next 5 years, we will see AIs such as ChatGPT grow and develop more abilities within, and potentially beyond, the domains of human intelligence. As they do, they will continue to change our world.
Some of the opportunities and challenges are known, and others are unpredictable.
We will likely see radical advances in science and medicine that will dramatically increase our life spans. Productivity increases will be measured in trillions of dollars. Radical advances in solar energy and even fusion are possible, collapsing the costs of energy and the goods that are produced with it.
Many job functions will be turned over to AIs, and this will hit “knowledge workers” the hardest. There may or may not be new, more fulfilling jobs to offset the old. Economies that depend on the sale of coal, oil, and natural gas may falter. Geopolitical shifts may emerge as wealthy countries are able to buy brainpower.
The value of the “knowledge” our students spend thousands of hours absorbing will radically decline, but a new emphasis on caring professions may emerge. Maybe the 2,000 kindest individuals will now be admitted to Harvard, and maybe fewer will complain about the admissions process if it creates a legacy of caring and kindness.
Last spring, I celebrated my 52nd birthday. When my own children celebrate their 52nd birthdays a little over 30 years from now, the world will be completely different, one that will be almost unrecognizable to many of us who will hopefully still be around. We will need a lot of care, kindness, and love to get there.