The Rise, Fall, and Rise of AI

After finishing grad school in the early 90’s, I ended up getting a job as the sole sales and marketing resource for a software company in Nanaimo on Vancouver Island. There were a handful of software companies on the Island at the time, but all of the others were involved in GIS (Geographical Information Systems) in support of the B.C. mining and forestry industries.

We were an AI firm. At the time, people would often ask you what AI stood for. Our focus was predictive modeling, and we used a variety of advanced technologies to build predictive models for financial, medical and process control systems.

This was the first bloom of industrial Artificial Intelligence. All this technology was coming out of academic and corporate labs, and people were trying to commercialize it. We bought a proprietary algorithm from an Idaho State professor. We purchased an expert system-neural network hybrid for trading futures contracts. For the tiny sector of the economy that we were involved with, AI was all the rage. Startups were starting, publications were publishing, and we were getting meetings. If you were talking about neural networks, fuzzy systems, expert systems, or genetic algorithms, people were interested in talking to you.

I met with a “quant” (although I don’t think that term existed at the time) at Bear Stearns. Motorola asked us to present at an internal conference on emerging technologies. I corresponded with a Viennese physician who was looking to predict blood glucose levels in diabetics.

It was heady times. The sky was the limit. Between the exuberance of youth and the intoxicating potential of AI, I thought our little company on Vancouver Island was going to take the world by storm. I told our CEO I was confident that we could sell 300,000 copies of the shrink wrapped version of our software over the next year.

Our sales a year later: 127. Unfortunately, neither our firm nor AI was ready for prime time. Some fuzzy logic went into appliances, neural networks ended up getting embedded into defence systems, and the rest went right back into the labs, not to emerge until about 5 years ago.

So what’s different this time?

The War is Over

For decades a debate raged as to what was the best approach to AI: rules-based or autonomous. For most of that time, it appeared that a rules-based approach had the upper hand. However, as time went on it became apparent that for the most promising applications (image recognition, predictive systems, natural language processing) autonomous systems were the only way to go. At a certain level of complexity, deterministic, rules-based systems were just overwhelmed.

Once this was decided, almost all the chips were tossed into the autonomous pot, and progress accelerated.

We’ve got the Power

One of the main reasons that the first AI emergence failed was that the technology was just too complex for the hardware of the time. Even with an SGI or Sun workstation, the volume of data to be processed swamped the processors (even the math co-processors!). Going through a round of training with a neural network would take a workstation days of number crunching.

Today, particularly via the cloud, the availability of processing power is virtually limitless. If you have the budget, you can deploy thousands of virtual machines as part of your AI project. This type of technology was unimaginable in the 90's.

A project that would have taken a year in the good old days, can be completed in three weeks today. On top of that, many projects can now be tackled that were not even possible in the past.

Hardware is no longer a bottleneck, and removing this hindrance has allowed AI the opportunity to blossom.

Is this a Bubble?

After twenty years, I’m thrilled to be involved in AI once again. Some of the meetings do remind me of that original AI bubble time, but in general everything seems more substantive now. Companies are spending real money on AI projects, and more and more of them are escaping from the Innovation Labs and getting operationally deployed.

It’s early, but I think the AI cat is permanently out of the bag. I do think things are going to move more slowly than a lot of the prognosticators predict. I think that most AI technology is going to be complementary rather than supplementary for a long time, but it’s all going to keep heading in the right direction. AI is going to make life better in a lot of different ways.

I don’t expect a robot apocalyse in my lifetime, or in my children’s lifetime. In fact, I don’t expect one at all.

Overall, in the AI world, everything is looking pretty rosy. Maybe it’s time to move back to Nanaimo and get the old gang back together!