2022 in AI, in verse and prose

There once was a year called “Twenty-Two”/ In which AI breakthroughs were brand new / We saw synthetics that think/ And robots that can link / To the world in a way that is true.

No, I didn’t write this limerick. Neither did I swipe it from anyone else. In fact, no human was involved in its composition. I had simply set out to write a year-end column on whether 2022 could be a turning point in artificial intelligence (AI) and turned to AI itself for help to get things going, having read far too much already about ChatGPT. This is the new AI-powered chatbot that has the chattering classes chattering like never before. It gives alarmingly well-composed, well-researched and well-punctuated responses to practically any question you can think of. I asked ChatGPT for a year-end sum-up of AI — in limerick form. You can judge the results: No Edward Lear, but serviceable.

I wondered if ChatGPT gets irony. I asked for an AI summation for the year, in the form of a joke. ChatGPT wrote back: Why was the AI researcher feeling depressed? Because he was having a breakthrough crisis! (I apologise if this joke falls flat. I am just a computer programme and do not have the ability to judge what humans might find humorous.)

Clearly, the stand-up comic industry has little to fear about being disrupted for now, but I would note that the AI is more self-aware than your uncle or school classmate who keeps sending you one bad WhatsApp joke after another.

How about a haiku? I got this: AI at new heights this year/ Synthetics that think and feel/ Quantum computing, wow.

My respect for AI, I must admit, was growing. At least it is modest. The “wow” was reserved for quantum computing, with no mention of the fact that ChatGPT itself has been the biggest wow-factor this year. But my reasons for getting impressed and worried at the same time increased further as I realised that the programme is good at faking it as well: Technically, it is trained on all internet data up to 2021 and doesn’t let on that it hasn’t caught up. No doubt, limericks, haikus and even the jokes will improve as its awareness of 2022 improves and it learns as more people use it. If it takes off, essays, legal documents, books, educational materials — even this newspaper column — can be assembled by a machine.

We may all struggle to tell if a composition comes from man or machine. You can see why ChatGPT has got the chattering classes in a stir as it is their jobs that would be among the earliest on the chopping block. Even the mighty Google is on Code Red: It is far more natural for people to ask a question and get a well-composed response from an articulate but extraordinarily nerdy assistant than type in search words and get lists of website links in return. The dethroning of Google search alone could make 2022 an AI milestone.

While “generative” AI, such as ChatGPT — and its cousin DALL-E, unveiled earlier this year, that produces images in response to text commands — stole the limelight, this summer, AI cracked the code to the very core of human existence — proteins, the building blocks of life. AlphaFold, developed by an AI research group, DeepMind, reported that it has predicted the structure of over 200 million proteins from more than 1 million species — nearly all proteins relevant to humans. The tool could revolutionise our understanding of disease and the search for new cures.

This development is especially encouraging considering AI’s performance during the greatest hour of need for AI. Earlier this year, I wrote an article in the Harvard Business Review analysing AI’s role in managing the surge of Covid-19 worldwide. Here’s what I found. The pandemic had created the perfect opportunity for AI to, literally, save the world. We needed fast, evidence-based decisions, there were datasets spilling out of every corner of the world and humanity was overwhelmed. For healthcare systems battling a novel, fast-spreading disease whose tests took too long to deliver results and a need to prioritise the most critical cases, AI was — in theory — the ideal tool. It could rapidly process vast amounts of information on symptoms and patient types and circumstances and make lifesaving triage decisions. Indeed, numerous AI tools were developed for this purpose. I found that across numerous reviews of hundreds of these tools, the conclusions were unambiguous — they failed miserably in practical clinical settings.

Now, failures themselves can prove to be prologue to breakthroughs. They lead you to ask, what went wrong? 2022 gave us an opportunity to reflect on why AI failed and consider the work to be done in the years ahead — breaking down barriers to the sharing of disparate datasets, constant iteration and learning through repeated use, fixing incentives of humans vested in the status quo, anticipating algorithmic biases and taking pre-emptive steps to overcome them and seeking AI talent more broadly from across the world and not just from a few mega-clusters, such as Silicon Valley, and a handful of elite companies.

2022 gave us a fascinating mix of AI goodies — a truly disruptive application that has already caused a stir, a foundational breakthrough that could mean a quiet revolution in the years to come and an epic failure that highlighted all that we must get right in 2023 and beyond.

To close out this year’s final column — in rhyme, of course — I turned to ChatGPT: AI breakthroughs at the year’s end as if Tagore had written it (in English translation, perhaps)? I got: As we move into the new year/ We wonder what AI will bring / Will it solve problems or create new ones / Only time will tell, with a single ring.

The writer is Dean of Global Business at The Fletcher School at Tufts University

Source link


By Google News

Google News is a news aggregator platform. It presents a continuous, customizable flow of articles organized from thousands of publishers and magazines.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.