We are used to think of AI as a futuristic technology from sci-fi movies, and at the same time we hear about AI being integrated more and more into our real world tools and devices.
It is pretty clear those two AI are not the same thing, in fact:
- Narrow AI is real, it’s here today and is a non-sentient and single-purpose form of AI.
- Wide, or General AI is not real or not yet real: at this point in time it only exists in sci-fi literature. It is normally described as multi-purpose, sentient, and often self-aware.
The idea of singularity, is normally associated with General AI, however I previously proposed the concept of microsingularity, as essentially a form of singularity that is related to a specific narrow AI.
Microsingularities -like Narrow AI- are very real and are already here today, examples are:
- artificial chess players (outperformed any human for a long time now),
- self driving cars (can outperform the average driver in terms of safety),
- detectors of false positives in cancer diagnosis (started outperforming human doctors recently),
- Natural language speakers (Google Assistant recently aimed at passing the touring test arguably: it was able to book restaurant tables and hairdresser appointments via phone without raising suspicion of being non-human).
As we gained access to narrow AI, and we have many working examples, it would seem possible to generate a wide AI by using a Narrow AI whose role is that of directing the information flow amongst different Narrow AI’s so that they would work together as one.
Let’s make an example of a taxi driver. He’s required to have multiple skills such as:
- knowing how to navigate the city
- making decisions based on traffic
- drive accurately avoiding accidents
- fine tune the driving experience according to the customer’s needs (“I am late”, or “I need to buy some bread”)
- adjust the environment to the customer’s needs (“it’s hot here” or “can we change radio station?”)
- chat with the customer is also normally quite expected.
We have several narrow AI’s as of this day that can do many of these tasks, but we do not have one single AI that can perform all of them as one experience.
A meta AI -an AI that governs AI’s- would be capable of doing all of these already today to a degree.
Today technology is mostly lacking in the chit-chat department, as we only took few limited attempts at the touring test yet, but essentially that is the one missing link: once voice assistants will be mature and capable of diverting voice commands to other sub AI’s, we’ll probably make a giant step towards a general AI.
So after all it is probably no coincidence that all major IT firms are investing so much efforts on different voice assistant projects.