If you haven't heard by now, there is a publicly available AI tool called ChatGPT. A lot of people are very excited by what it can do, and not just people, companies too. For example, Microsoft and Google are practically tripping over each other to try to bring a tool like ChatGPT to market as an assistant to both search and their productivity and document creation apps.
If you've never heard about any of this before, I'll try to get you up to speed with what has been developed so far. If you already know about these AI tools in their current form, you may as well skip a paragraph or two. ChatGPT is a natural language AI tool developed by a company called OpenAI. The tool was trained with an extraordinary amount of information available online in every topic, subject, and discipline, including computer programming. The tool itself was then designed to provide an answer to any plain English question in the most helpful manner possible. You can ask it for a type of recipe, and it will give you one. You can ask it how to create an Excel formula given very specific parameters, and it will give you that formula. You can ask it to generate a stand-up routine on a specific topic in the same style as a known comedian, and it will generate one. You can ask it to generate a three page essay on the benefits of universal basic income, and it will give you that essay. In an amusing, yet fascinating turn, you can then ask it to grade that essay with the critical eye of a 20-year tenured English Literature professor. It will then proceed to tear that essay apart.
But this is only scratching the surface. Many people are only trying this tool using two-dimensional thinking. In other words, it's one thing to ask the tool to help you with choosing a new career path, it's quite another if you prepare the tool by instructing it that it is a seasoned professional career advisor who has been advising people on career transitions for over 20 years, and prompting the tool that before offering any advice, it must ask a question first to focus the conversation on what the user wants to discuss. It really is something to behold. It is not perfect, however. A friend told me that he had asked it to generate a string of numbers in a Fibonacci sequence and it got some of the numbers wrong. Only when he instructed it to perform the same task, but show the proof of how it arrived at the result did it get the numbers right. Another time, I asked it to list all the years that the Toronto Maple Leafs lost in the Stanley Cup playoff finals and it gave me a list that wasn’t accurate. But then I realized that it had misinterpreted my query to mean ‘all the years that the Leafs lost in the playoffs’, period. Not just the final series. Then the list made sense. As a good friend of mine put it, “ChatGPT is like an autistic savant.” It knows a lot of stuff, but it sometimes has difficulty answering your question.
But this new AI reality isn't just about answering questions and solving problems. There now exist tools that will generate new art based on whatever you ask it to create, as specific as your instructions. So if you asked the tool to generate a picture of a human being, it will decide the gender, the pose, how much of the body is in the frame, what is in the background, the style of the image, etc. But if you ask it to create a picture in black and white, of a lumberjack standing in a dense forest, holding an axe, in the style of a horror movie, that's pretty much what you will get. The results may be extraordinary, or downright bizarre, and if you don’t like what you get, you can tell it to try again. As you begin to recognize what prompts produce certain results, you can generate some mind bogglingly superb art. Pictured below, is an AI generated image based on my prompt, "Two robots arguing with each other and one of them is emanating smoke."
This has some people quite worried. There is a potential for students to cheat by asking an AI tool to perform work on their behalf. There is a potential for workers to ask an AI tool to perform work tasks on their behalf. Journalists have already proven that the tool can be used to generate entire news articles on a particular subject, so long as the AI is aware of and has been trained on the subject in the first place. Early adopters even tried to use the tool to instruct them how to do unethical things. As a result, the folks behind the tool’s creation had to implement ethical blocks to prevent misuse, which clever human beings have found a way to circumvent in certain circumstances. The point is that this tool, like any other tool, can be used for good and bad. For example, if a university student just isn't quite getting a specific topic in chemistry, it could ask the AI tool to explain the topic at the level of a high school student, and that's exactly what they will get. I love this particular aspect of the tool, because I love using analogies to explain complicated topics, and ChatGPT is superb at explaining things at the level of a child, which then inspires me to come up with my own analogy that I can call my own.
Lately, a colleague of mine showed me how he was using an AI art tool to ‘suggest possible minimalist icons for an application brand name featuring a bee’, and lo and behold, the stuff it came up with was incredible. Absolutely astounding.
The possibilities truly are endless, and keep in mind that this is just the first phase of AI tools. Within months of this writing, the tools will have evolved to be more capable, more accurate, with better context. The people behind the magic have planned specialized tools that will help diagnose health symptoms, add helpers to your productivity software to make your documents more readable, generate presentations based on existing documents, find bugs in programming code, analyze architecture designs for structural strength, and so much more. If you trained the AI on the entire tax code for a country, it would potentially do the absolute best job of figuring out your tax return to get the most deductions while remaining legal. An AI could potentially tell a diabetic how much of a type of food they could eat, given their current blood sugar level, without going over a certain threshold.
The question has been posed, “How are students going to learn anymore when they can rely on an AI to give them the answers that they need?” Simply put, schools are going to need to change their approach to learning and testing. The same problem arose many years ago with the invention of the pocket calculator. But a student could also rightfully ask, “What is to stop my teacher from using AI to grade my test in a manner bereft of empathy and humanity?”
The reason why I am excited about this technology, is because the Internet has made it possible for us to access an awful lot of information, but the existing search tools haven't really made it possible to access that information without us having a full understanding of a search site’s syntax. I have seen numerous examples of this, when I speak to people who can't find something online, yet I have no trouble finding it using a better combination of search terms. In other words, access is not equally available to all participants. AI changes this, by making it possible to get answers using plain English queries. We already live in a world where a growing number of people don’t say the words ‘I don't know’ very often. AI is going to make it possible to eliminate that phrase from our vocabulary forever.
No comments:
Post a Comment