I’ve been very fortunate to witness two huge technological disruptions in my life. Microprocessors came onto the scene when I was a young teen. Then in 1993, the Internet arrived. And now, this year, an Artificial Intelligence (AI) revolution is upon us.
This Time It’s Really Real
I worked in AI as a young intern in the mid-1980s during the first AI boom, which soon turned into a bust. Computers weren’t powerful enough at the time to do anything too useful.
But ten years ago, Graphic Processing Units (GPUs), which had been built to draw 3D graphics for real time computer games, were discovered to be a great tool to also run a type of AI program called a Neural Network (NN). Since then, GPUs and NN architectures have grown in lockstep with each other.
Your phone’s face recognition unlock feature is powered by AI, as is Siri’s voice recognition function.
A few years ago, dedicated AI chips made their way into the mainstream, no longer reliant on gaming GPUs (each iPhone has an AI chip), and a new NN architecture called the Transformer hit the scene, allowing very sophisticated AI systems like Tesla’s self driving car feature.
But this year, two giant leaps were made in two very different areas that made everyone sit up and take notice.
Generative AI Art
In September, an AI “artist” won first place in a digital art competition.
It was created by a “text-to-image” AI program called Midjourney. With Midjourney, you ask the AI to create an image using an English language sentence. For example, the stunning image below was generated by asking Midjourney: “beautiful, fantasy city unreal engine”. Unreal engine is an artistic style used in computer games.
I’ve been using various AI art generation programs (like DALL-E) myself over the past few months, and many recent articles here in the Post are using AI generated artwork for the title image (including this article).
At the very least, text-to-image AIs will dramatically improve commercial artist productivity. But text-to-image is only the beginning. The race is now on to create text-to-video. There are already a number of companies that will create your next training video, or corporate presentation video, complete with AI generated photo realistic talking heads. All you need is a script and the AI will speak the words and create a realistic human talking head.
But beyond even this, creating quality video animation or special effects for movies is a hugely costly endeavor today. It doesn’t require a crystal ball to see that this technology will eventually allow text-to-animation putting huge numbers of animators and even filmmakers out of business.
Ah well, at least creative story writing is safe, right?
ChatGPT: Oh My God!
December 1st rolled around and a non-profit AI company, OpenAI, released their latest “Large Language Model” called ChatGPT.
It can write stories for you. It can answer complex questions about technical subjects in an easy to understand way. It can carry on a full conversation with you including past context sounding every bit like an intelligent, knowledgeable human. You can paste a complex scientific paper into it and it will summarize it for you. It can write your next high school and even college English essay for you.
These are not exaggerations! In this section I’ve got examples, but you have to try it out yourself.
> In the Book 1984, what society is George Orwell portraying?
A dystopian society in which the government has compete control over the lives of its citizens. The book is a warning about the dangers of totalitarianism and the importance of individual freedom and autonomy.
> But at the time of its writing, was there a particular society that the book was warning us against?
It was a warning against the rise of authoritarian regimes in the wake of World War II. At the time he wrote the book, several countries were ruled by authoritarian regimes, including the Soviet Union, Nazi Germany and Fascist Italy.
Some creative stuff it produces can be pretty funny:
> Write me a funny scene in a play where the homeowner’s association is trying to force a homeowner to do something silly and absurd?
(The scene takes place in front of a house. There are three members of the homeowner’s association, all wearing suits and carrying clipboards. One of them, the leader, is speaking.)
LEADER: Mr. Johnson, as you know, the homeowner’s association has very strict rules about the appearance of houses in this neighborhood.
MR. JOHNSON: Yes, I’m aware of that.
LEADER: And one of those rules is that all houses must be painted in a cheerful and pleasing color.
MR. JOHNSON: Okay, I understand. My house is currently painted a nice shade of blue. Is that not pleasing enough?
LEADER: I’m afraid not, Mr. Johnson. The homeowner’s association has determined that your house needs a brighter, more eye-catching color.
MR. JOHNSON: What do you suggest?
LEADER: We suggest you paint your house pink.
MR. JOHNSON: Pink?!
LEADER: Yes, pink. It’s a very cheerful and pleasing color, and we think it would really brighten up the neighborhood.
MR. JOHNSON: (incredulous) I can’t believe you’re suggesting that I paint my house pink. That’s ridiculous!
LEADER: (sternly) Mr. Johnson, the homeowner’s association has made its decision. If you don’t paint your house pink, we will have no choice but to impose a fine.
At least, I thought it was funny.
The truly scary thing is that ChatGPT is just a research prototype that the company put out to learn more about it. It wasn’t even meant to be a useful productivity tool, and yet people are using it for all kids of real world applications.
It writes computer code! I asked it to write a program in Python to factor a number, and it did it! Well commented code too, and an English description of how the program worked. And then I asked it “for the same program but in PHP”, and it wrote that right away along with a description of how it was different from the Python one!
To be sure there are limitations. ChatGPT doesn’t reason very well (ask it which is heavier, 10 pounds of lead or 10 pounds of feathers, and it is hit or miss if you get the right answer), and if you ask it to write a paper with references, it will happily oblige, but the references will literally be fabricated. ChatGPT isn’t connected to the Internet (anymore at least, that’s how it learned things in the first place), so it can’t do a current search for you or interact with the real world on your behalf.
But that’s an artificial limitation intentionally put in by OpenAI. Future “chatbots” won’t be so constrained. Indeed, you.com is a search company that has integrated web search with a large language model. People using ChatGPT remarked how much more useful it is than Google, and you.com has jumped in and essentially merged the two functions. You.com provides answers in English text and then provides links to more comprehensive information.
Predictions Are Hard
If you asked computation experts ten years ago how AI would eventually affect society they would have said something to the effect that first, blue collar jobs would be affected, then lower paid white collar, and it would eventually work its way all to the way up to top flight scientific jobs and finally fully creative jobs like artists.
The exact opposite is happening.
Deep hard scientific problems like protein folding have been solved by AI, something that maybe no human could have solved. More recently, fusion energy got a boost when an AI was able to predict plasma instabilities and actually contain a fusion plasma in real time, something heretofore that had been an intractable problem.
And the generative artistic AIs are winning digital art competitions without even breaking a sweat. Both creativity and artistic talent have fallen by the waysides in a stunningly short time period.
Meanwhile, blue collar jobs appear, for now, to be safe. So, predicting where this technology will go, which is just now hitting us, is probably a fool’s errand. But I’m just that kind of fool to try, so here it goes.
Where Are We Headed?
It appears that advanced AI is impacting knowledge worker fields first. Scientists, engineers, artists, computer programmers and writers aren’t being replaced (yet), but by using these advanced AI tools, their productivity will soar.
This will take many years to play out. The Internet became a thing around 1995, and twenty years later, it was still being integrated more fully into society.
Venture capital money is being poured into startups and mature companies alike that are building tools for various business verticals. Everything from doctors to script writers are going to see tools that incorporate advanced AI to help them perform their jobs better. Examples:
- Harvey for legal research
- Runway for commercial art
- Github Co-Pilot for AI generated code
- Cradle for AI generated proteins
- Synthesia to create AI generated training and marketing videos
Big advances will occur in life sciences, which has no shortage of progress lately, but it will probably get even crazier. Certain types of cancers are already being cured. More will follow along with cures for many genetic diseases.
ChatGPT functionality is going to be rolled out everywhere. The base large language model will be augmented with specific knowledge in whatever field (legal research is soon going to get a make over!) and sold as an Internet service to various knowledge workers.
The Robots Are Coming
Robots, of course, have been here for decades. Watch any modern day automotive assembly plant (video) and you’ll see a large number of industrial robots welding frames together. But this doesn’t require AI since the environment the robots operate in is very regular. They are programmed to do the exact same repetitive thing over and over. Not much intelligence is being used.
Robots in the outside world are a different story. Here, nothing is ever the same, and a robot has to first perceive its environment using sophisticated vision processing. Advanced AI is starting to be used here. Expect to see interesting robots anywhere from soon to 5+ years from now.
In the “soon” category, look no further than Tesla with its “Full Self Driving” (FSD) software add-on for their cars that actually drive everywhere. Urban, freeway, at night, rain, in the mountains, etc. It is not perfect yet, but I’ve been using it myself and do about half my driving with it (it is cautious and more slow than I’d like when I’m in a hurry).
In the 5+ year out category, we will see mobile robots doing flexible work assignments, replacing the last frontier of work, blue collar jobs.
Google has a research robot division called Everyday Robots where they are currently combining a Large Language Model with robot behavior and planning (video).
Meanwhile, Tesla again, is building a humanoid robot that will leverage their vision perception AI they have built for their vehicle driving system (video).
Conclusion
I distinctly remember trying to build a robot when I was eight years old. I quickly came to the conclusion, even then, that the hardest part (or at least the part that most interested me) was the brain. And I was shocked to realize that no one knew how to build a robot brain. I’ve been on the lookout for fifty years waiting for technology to come together to make such a thing a reality. We are now almost there…
Addendum, here is an incomplete list of AI product companies today:
