The following was given as the intro to a discussion panel for teachers Coram Deo Academy. It has been changed slightly to fit the blog post format. Thank you to Jon Jordan for the invite!
Struggling with AI
Last semester I taught a computer programming elective for middle schoolers, and the entire semester I struggled with AI. The platform we used for coding decided to move away helping kids to write code by hand to go all-in on AI and have them prompt their way through it. On their homepage it asks “what app do you want to make?”. They added so many AI features that was difficult to completely turn off, so I spent significant time in class making sure that every AI helper was turned off on every student’s account.
Thankfully, it is very easy to decipher code written by AI and code written by 12-year-olds.
But on the last day of class I gave in. Since it’s undeniable that AI is changing the art of programming, I decided to have the students build apps with AI tools so that they were familiar with the future of programming.
The results were astounding. The kids that struggled the most to understand programming concepts suddenly could unleash their creativity to create the apps they’d always wanted. In 30 minutes, one student that struggled with programming the entire year created a 3D game with levels, characters, and the beginning of a storyline.
And he had no idea how any of it worked.
Humans have told stories of talking inanimate objects for millennia. And for the first time in human history this dream has come true, through complex arrays of silicon circuits, software, and super computers. Not only for a select few, but for just about anyone. It’s getting cheaper and gaining new capabilities incredibly quickly.
Before we can understand what the future holds with AI, we must first define what it is and what it is not - which I’ll do my best to tackle.
AI is a Technology
First, AI is a technology. That might sound obvious, but it’s important because all technologies solve some problems, and they also create brand new ones.
Just like the wheel, or eyeglasses, or internal combustion engines, AI is a useful technology for many human purposes by making certain aspects of life easier and allowing us to do things we never could do before. I am personally seeing this change in professional software development, where full apps are created in days instead of months by those with little experience programming. Proponents of AI in education tout the benefits of generated lesson plans, automated grading advice, and assisting students learn difficult concepts. Stories abound of just how good AI is at these tasks.
But new problems are always created by new technologies. The internal combustion engine revolutionized transportation, but also created the environment for automobile accidents, pollution, oil spills, traffic, breakdowns, oil changes, and drunk drivers. In the same way, we are seeing new problems arise with AI. AI makes it much easier to cheat, and harder to get caught doing so. Spam is getting much worse as AI farms crank out hyper-personalized messages. It is harder to trust pictures or videos shown on social media, and easier for hackers to gain access to private data. A proliferation of AI girlfriend or boyfriend apps promise relationships with all the perks and none of the downsides. Stories of clinical grief are arising when an AI significant other is accidentally deleted.
AI is a Simulation
All this leads to my second point. AI is a simulation. This is important because many believe otherwise. It is now common by Silicon Valley elites to refer to a person’s brain as a “biological neural network”, and to talk about a future where there will be silicon-based lifeforms at some point instead of just carbon-based lifeforms.
A simulation, by definition, is not the actual thing. The simulation is always derived from something else and is not its own entity. Simulations do not experience consequences like the real thing its mimicking. Which is in fact the reason why many things are simulated – for example car crashes are simulated on a computer because a fake car can be crashed as many times as necessary and the design tweaked accordingly.
When it comes to AI the same is true. A simulated human may seem to have many of the same thoughts, emotions, and reactions as a real human, but it can never experience the weight of the consequences of its decisions. If a person tells an AI to invest their money, and the AI loses it, the person is the one that feels the pain. If an AI breaks a law, only the people in charge of the AI or the organization that created it can be held accountable. This means that there is a limit as to what AI can oversee - even if it becomes incredibly competent - because important and risky objectives will always need a real person at the helm.
Simulations Cannot Tell Real From Fake
The other thing about simulations is that there is no way for something simulated to know what is real or fake. This is because a simulation always has something else sensing for it. There is a medium in between its senses and the real world. In the case of AI, it is trained only on what its trainers give it. It cannot investigate the real world itself. And even if it could – say if someone invented a robot that can learn like a human – there is always the chance someone could be in the middle modifying the inputs.
This means is that AI cannot discern truth.
AI has no way to determine whether it’s perceiving the real world or if there’s someone behind the scenes toying with things. If an AI cannot discern what is true, then it will always be gullible based on the wishes of whomever is controlling it – for good or for ill.
Only Humans Can Discern Truth
AI is a powerful tool that will reshape the world. In the coming years it will become even better, break records, and do things thought impossible. At the same time, there are fundamental limits to what it can do. Our objective as educators is to help our students learn the art of discerning truth for themselves. That will continue to be a task only humans can do.