Let’s start with the disclaimer: we’ll talk here about Artificial Intelligence (AI) in its least glamorous form. If you’re aiming to Skynet or Terminator, I would suggest starting from bipedal robots that are struggling to make their way through a closed door. As it turns out this is a huge task for modern robotics, including Boston Dynamics. Our focus point will be more primitive and lean: what we call it – “super-weak AI”.
Classification of AI systems based on problems they solve
Taxonomy of AIs considers two main categories: weak and strong. Weak AI is designed for very narrow tasks. This type of AI can’t be used for any problem besides it designed for.
An application that sorting out cats from dogs is a good example of weak AI. If that hits the rat, for example, it’s likely to classify it as a dog. This is the limitation of weak AI: it can only deal with data that it’s trained for. It is not true intellect and cannot generalize, accumulate experience, or be sensitive to the context.
When it comes to consciousness and rationality: it is not even agreed from a philosophical standpoint among us, humans. Still, the idea of strong AI is based on the ability of an AI to embrace these qualities.
For now, strong AI is more myth than reality. On contrary weak AI, found its path to the client. We see it being implemented in personal assistants (Siri, Alexa& CO), cameras, NLP, and image processing.
Weak AI is already being used to solve simple tasks. Strong AI requires solving philosophical problems of consciousness, self-identification, values, and contextual memory. It’s still a rather futuristic concept.
The third type of AI can be named super-weak AI. It is very primitive, extremely limited for the purpose and works only in combination with similar kinds. Its main feature is the ability to perform a very small amount of highly specialized analysis in the shortest possible time and share results to neighbors. Sounds familiar? Yeah, say, ants or bees.
That’s all good, but do we really need super-weak intelligence, and if so, why?
Where weak AI fails, super-weak AI can succeed. How is that possible?
Current AI solutions encounter many problems. Take an Autonomous Driving for example. It has to process large amounts of data coming from sensors and video, limited by computing power and energy supply, and must act fast.
Every car that is moving in the traffic has to do this job. It may rely on other’s car data but how to make sure that data is reliable and neighbor vehicle wasn’t compromised by hacker. Hence, full attention to the road and dull functionality at the end.
ADAS drives carefully, assuming the worst-case scenario. But if something actually goes wrong, the car suddenly seeks help from the driver. With this level of AI, the driver can only dream of reading or watching a movie behind the wheel. What about tasks that require more focus stolen from driving itself? The solution can be found in the previously unexplored terrains – super-weak AI hurries to rescue.
So what can super-weak AI offer? Instead of performing all calculations itself, a car using super-weak AI can rely on the data from other cars and close infrastructure. Super-weak AI function is limited to critical situations evaluation while rest of its power dedicate to data verification that it received from the network.
Finally, the car itself loses its ability to move fast alone, but get an ability to move faster when there are more similar AIs are hitting the road. And in the long run, this is quite possible since the percent of AI featured cars is rising fast.
Estimated Growth of AI featured cars
Source: Business Insider
If super-weak AI is so cool, why is no one using it?
First of all, there aren’t enough autonomous cars on the road and, more importantly, there is no trust can be found in such highly distributed systems.
The solution for trust problem is decentralized consensus, that can be found in neighboring hype galaxy – the blockchain.
It turns out that AI and the blockchain can be merged and achieve lots of benefits. First, it can build local consensus based on multiple data principals and constant verification of events exchanged within a network.
For a super-weak AI is critical to focus on small separated tasks. Every node should work together on the common goal. Within the properly selected network architecture AI can reach multiple goals: discover critical objects on the path and verify neighbors’ data. Hence, the combined computing power of the nodes will maximize optimization value of the system by dropping individual priorities and build trustful communication environment.
What would super-weak AI look like?
Let’s imagine we’re in the year 2050. Most cars are autonomous. They actively exchange data with each other. If some starts pushing its wrong (say compromised or attacked by old-geek manual driver) the whole traffic determines this object, reduce its trust level and avoid it.
Road infrastructure (traffic lights, bumpers, markings) has rudimentary intelligence. Infrastructure collects data and informs passing cars about road conditions. The whole traffic starting to move as good dancers on the floor.
Cars go faster than they do today by calculating the safest distance and trajectory based on the current situation and data collected from everywhere.
At a speed of 100 kilometers per hour with a decision-making frequency of 10 times per second, the safe clearance between cars reduced to six meters. Side cars can be moved 10 centimeters away. Compare this to current 150 meters front-to-back and 3 meters on the sides and push hard your imagination. Nobody waits on traffic lights. Traffic jams are passed to the history.
This condensed traffic pattern quadruples the capacity of existing road infrastructure. It expands areas for sidewalks and parks.
The economic effect of full-scale autonomous infrastructure is difficult to imagine. According to recent numbers, the US would need to spend $2 trillion over the next 10 years to repair roads.
US infrastructure needs over the next 10 years
Systems that use weak AI have become part of our everyday lives and are gradually gaining wider use. The development of strong AI will have to wait for advancing the problems of consciousness, rationalization, and generalization. As for super-weak AI, it can be a breakthrough for many sectors including autonomous driving.
If you’re interested in AI and concepts above, please, contact our experts.