When it comes to the topic of Artificial Intelligence in general there are many interesting opinions the general populous holds, as well as misconceptions. The biggest of these by far is the understanding of what Artificial General Intelligence will look like. The potential of AGI cannot be overstated and I mean that from the stance of mathematical impossibility. The world and human history as a whole will be forever changed when we reach that rapidly approaching threshold. For the scope of this article though I am not here to simply write about the marvel of what may be humanities’ last creation. That milestone event is still some time distanced from the present at time of writing. I am instead here to pen my thoughts on the predecessor of AGI that is on the verge of being created: Proto Artificial General Intelligence.
As the remarkably unclever wording suggests, Proto-AGI is a less advanced version of AGI proper. To start with a simple definition, a Proto-AGI is an AI that is capable of a wide variety of tasks without prior input on how to do them. In many ways it resembles the human ability to use reason, head knowledge, and situation to correctly guess how to interact with the environment around them. Notice how I said it resembles the human ability. It is not actually reasoning. It is not conscious and certainly not sentient. It is however unlike a normal neural network or AI.
In normal AI development the process looks something like this: take a task such as identifying photos of cats, take a database of millions of cat photos and feed it to a neural network, go through testing it and telling it if it guessed correctly or not, retrain it, and repeat until it does the task to a high degree of accuracy. Sure, the AI is guessing something correctly. It is not however “smart”. It cannot identify and label other images. All it can do is what it was trained to do: tell if an image has a cat or not. Past that singular use the algorithm is useless.
Proto-AGI are different. Unlike the typical model they can be trained on multiple things. For example, you could train it to generate music, video, and text. The same AI, but doing the tasks of what would normally be three. Such AI would have vastly more potential simply because of their utility. You could have a movie theatre run by the same Proto-AGI split across devices. Some would be put into robots that clean and cook. Another would take tickets at the front and sort by movie. Still another could handle the accounting and running of the business. Finally one could even dynamically generate a movie based on audience input in real time (something I have predicted in another post that we’re already extremely close to). Same AI model, told to do different things. It is given the command and will use what knowledge is relevant to the completion of the task provided.
The more thought you give to the example above the quicker you will realize just how vast the potential is. We don’t need AI that is conscious to do untrained tasks. In fact a Proto-AGI could do virtually all tasks given to it once it is sufficiently advanced. It could automate factories. Design better materials. Plan the logistics of an entire nation. The list goes on. This is the technological marvel we are tantalizingly close to unleashing. How will we deal with a future where every job can be automated? That’s up for us to decide. I will cover that more in a future article.