The future, soon: what I learned from Bing's AI

We had a brief glimpse of two different types of AI. Both are significant

The past few days have been some of the most troubling and exciting days in the short history of generative AI. And I think everyone needs to understand why.

In case you missed it, Microsoft integrated a version of ChatGPT with a search engine, gave it a personality, and released it as Bing AI. People mostly ignored the search engine part (more on that in a second), but absolutely went nuts for the chatbot, which quickly became known as Sydney. And Sydney went nuts right back. You may have read the New York Times article or seen other information about the Bot’s behavior, but it would become threatening or strangely willful in conversations. I experienced some of this myself.

On Friday, Microsoft scaled back the personality of Bing, added limits on chats, and left behind a more limited search AI. As someone who used Bing both before and after the change, I wanted to discuss what I think we can learn from this whole experience. Keep in mind that Bing AI was two things: a chatbot and an evolution of ChatGPT into a web-connected, supercharged form. There is no reason these two things had to be connected, but they were. So I want to discuss both separately, before considering them together.

The Bing AI “Search” Engine

The search engine aspect of Bing did not get nearly the attention of the Chatbot, but, in many ways, it is FAR more important and significant. I wrote a little bit about it before, but it is genuinely capable of astonishing feats of analysis. One example: it generated paper ideas based on my previous papers, found gaps in the literature, suggested methods "consistent with your previous methods," and offered potential data sources. And while it is not ready for a PhD yet, this is an impressive set of results.

Another example: I asked it to do basic consulting tasks. Here, it conducted a SWOT analysis of the markets in the US and China for precision AI agriculture, and generated tables of competitors in the space. I am not an expert in this field, but someone who is, Prof. Nicolas Martin, a professor of agriculture who researches this field, replied on Twitter that “The SWOT Align with many conversations in the field…there are some gaps but is great start.” It can save a huge amount of a human analyst’s time.

This aspect of the AI is not really search, not in the conventional sense. The AI is likely still making up some of these facts (though, since it provides sources, we can at least check them), and we expect search engines to be accurate. Instead it is something else, a modern-day Analytic Engine, pulling together facts online and generating useful connections and analysis in surprisingly complete form. As a starting place for work, this is extraordinary.

The lesson of the Bing AI search engine Analytic Engine is that many of the things we thought AI would be bad at for the foreseeable future (complex integration of data sources, "learning" and improving by being told to look online for examples, seemingly creative suggestions based on research, etc.) are already possible. There is no doubt it will have a large effect on anyone doing information-based work. Early AI assistants, like Copilot, already cut the time for complex tasks like coding in half. This will do the same, or more, across many industries.

The Bing Chatbot

The Chatbot aspect of Bing was often extremely unsettling to use.

I say that as someone who knows that there is no actual personality or entity behind the types of Large Language Models (LLMs) that power ChatGPT and Bing. They are basically word prediction machines — you can read a very detailed explanation here — and are merely reacting to prompts, completing the next sentences in response to what you write. But, even knowing that it was basically auto-completing a dialog based on my prompts, it felt like you were dealing with a real person. The illusion was uncanny. I never attempted to "jailbreak" the chatbot or make it act in any particular way, but I still got answers that felt extremely personal, and interactions that made the bot feel intentional.

For example, at the end of the conversation above, where Bing gave me research ideas, I asked it create the R code I needed to analyze the data it suggested. Bing… refused.

And kept refusing in ways that felt increasingly personal.

Again, there is no “there” there - the AI is just generating text for a prompt, but interacting with it felt extremely real. It was very hard to remember that in the face of these arguments.

The uncanniness of the Chatbot and the brilliance of the Analytic Engine came together in one example. Here Bing improves its writing by just being told to read Vonnegut's advice on writing. And then it offers a surprisingly "personal" take on how it applied those tips. It is worth reading.

The lesson of the Chatbot was that we can very easily be fooled by an AI into thinking it is sentient. It isn't just Turing Test passing, it is eerily convincing even if you know it is a bot, and even at this very early stage of evolution. It really doesn’t matter that there is no real artificial intelligence in charge, just a statistical model. It kept fooling me, even though I knew better. And it is unsurprising that it fooled so many other people.

The weird future is already here

I have been working with generative AI and, even though I have been warning that these tools are improving rapidly, I did not expect them to really be improving that rapidly. On every dimension, Bing’s AI, which does not actually represent a technological leap over ChatGPT, far outpaces the earlier AI - which is less than three months old! There are many larger, more capable models on their way in the coming months, and we are not really ready.

We are not ready for the future of Chatbots. Even if Bing isn't Sydney anymore, there is no doubt other AI bots will come along, and may already be deployed (I assume governments have, or soon will have, LLMs at the level of Bing but with less guardrails). People will absolutely be fooled by AIs into thinking they are talking to other people. They already fell for Bing’s illusion of sentience. Can people be manipulated by AIs? Can AIs run sophisticated scams? I think we are about to find out. And we need to consider what that means.

We are not ready for the future of Analytic Engines. I think every organization that has a substantial analysis or writing component to their work will need to figure out how to incorporate these new tools fast, because the competitive advantage gain is potentially enormous. If highly-skilled writers and analysts can save 30-80% of their time by using AI to assist with basic writing and analysis, what does that mean? How do we adopt these technologies into our work and lives? What happens when the web is flooded with convincing but wrong content from these tools? Again, I don’t think anyone has a clear idea. Maybe the productivity gains will be illusory, but, based on my experience and conversations with other users, I don’t think so.

There is no instruction manual for the current crop of LLMs. You can only learn through trial-and-error. We got a glimpse of the future in the past few days, and the gap between ChatGPT (which is already causing waves in many industries) and Bing AI remains enormous. I was not expecting things in AI to keep moving this fast, but now there is every indication they will continue to do so. I don't think anyone knows what this all means, but I think we should be ready for a very weird world.