• Technology

A complete beginner’s guide to AI Assistants

Several major advances in machine learning and AI assistants happened within recent memory. The one I heard about first was Midjourney, which could generate artwork using text prompts. Then I began hearing about Open AI’s ChatGPT, which initially sounded like a really convincing chatbot. But now it’s six months later and I’ve read about people having it write code, plan meals, create travel itineraries, and more.

Since I love science fiction, I’ve spent a lot of time ruminating on things like artificial intelligence, the nature of consciousness, the possibility of the singularity, and how our lives would change with extremely advanced technology. This was the first time I started to actually worry my livelihood might be affected.

Some important background

The way computers have been programmed in the past requires explicit instructions called algorithms. These are usually written in programming languages or data that’s easy for a computer to process. A basic example of an algorithm would be a decision tree, where the computer goes through a set of “if this, then that” logic to arrive at a conclusion.

Part of what makes humans, plants, and animals resilient is our ability to learn without being explicitly told things. Speaking broadly, we take the world in through our different senses, adapt, and flourish as best as we can. When we arrive here we’re hardwired to do a lot of the things we do automatically. There’s no need for a plant to have a “photosynthesis algorithm”, it just photosynthesizes.

Computers are the opposite. They’re blank slates until we fill them up with algorithms and information to process. Throughout the years people have attempted to make computers behave more like humans, but it’s often in extremely narrow and limited circumstances compared to those that a human faces. I’ve played video games where computer-controlled characters behaved in really convincing ways, but that was within that specific game’s narrow confines. In order to act as a convincing human within our reality, it would presumably need a hard drive big enough to hold the same amount of information as we’re able (apparently 2,500 terabytes?), algorithms and sensors that simulate all of our senses, a processor fast enough to take everything in, and the time to learn everything we learn in our lifetimes.

Well, we have some shortcuts we can take and the two you need to know about are machine learning (ML) and natural language processing (NLP).

What is Machine Learning?

Machine learning is a technique that enables computer programs to learn and improve from experience without being explicitly programmed, similar to the way organisms do. But instead of existing for a long time and slowly taking life in, machine learning sends large amounts of information to complex algorithms which allow the computer to adapt on its own.

This technology has been in use in various industries such as healthcare, finance, and transportation. But AI assistants are the first software programs designed to perform tasks or services for average people.

What is Natural Language Processing?

Natural language processing (NLP) is a subset of machine learning that focuses on teaching computers to understand and generate human language. It involves training algorithms to recognize and respond to human language. NLP is how your phone’s assistant figures out what you mean when you ask it for something, and it’s how AI assistants

What are some of the leading AI Assistants?

ChatGPT, Bard, and Bing are the big three currently. If you’d like an in-depth breakdown of the pro’s and con’s of each, I recommend watching this YouTube video by Marques Brownlee. They’re presently all free, but collect information from you to make their products smarter. Of course you likely have Google, Siri, or Alexa at the ready as well. The big difference is those are voice operated and a little more limited in scope.

How smart is AI?

Like the convincing video game character I talked about earlier, AI is still working within the confines of its algorithms and the data it’s been fed. The biggest tech companies in the world are betting big on AI because it has immense potential that could lead to us solving some of the world’s biggest problems. But it’s not quite there yet, and that tends to show the most with AI assistants. Sometimes they give inaccurate information, or just give weird responses. The data and algorithm need tweaking, and this is where things get interesting and a little scary.

Algorithm Synthesis

There are machine learning engineers and data scientists who manually make adjustments. But machine learning itself can create new algorithms that it can use in a process called automated algorithm design, or algorithm synthesis. This is effectively the ability for a computer program to learn from experience, self-improve, and use that in the future. It’s an emerging area of research, so it’s anyone’s guess as to how big of a deal this will really be, but the potential is enormous.

So uh… should we be worried?

If you’ve seen Terminator 2: Judgement Day, you’re familiar with the idea of the technological singularity. If not, the hypothesis is there will be a point in the future where the technology we create advances to a point that humanity loses control over it, and it turns into a “runaway reaction” of self-improvement that breeds an artificial superintelligence (ASI.) There are a lot of smart people who have debated the plausibility of this happening, as well as the likelihood that it would be friendly to us. Folks who believe this is a reality we need to be paying attention to speculate it will happen in the next 10–30 years. Other people believe there’s a complexity brake which will push this back substantially or prevent it entirely. The reality is we don’t really know if or when this scenario would happen, or if we should be worried about it. But even if we knew for certain that this were on the horizon, there’s little any single person can do to stop it or prepare for it. Similar to wringing your hands about global nuclear war, worrying about it doesn’t do much good.

But should we be worried about our jobs and our livelihoods with computational power this advanced?

Even for how advanced these new tools are, on their own they still can’t replace the human element entirely. Just like search engines and the Internet were new tools at one time, the ways we do our jobs are what’s actually changing. The tools won’t replace people who don’t use them, but people who use the tools will replace the people who don’t. (Incidentally, about two paragraphs of this entry was written by A.I.)