AI at work
I asked a story writing AI (www.deepstory.ai) to create this introduction by feeding it with this text:
“Self-driving cars are the future but they need much more work yet, and the AI needs to improve…”
The suggested addition was:
“…to make them more appropriate for the future, and the question that needs to be asked is how much more work is required for what might be our future? Well, that question does not have a simple answer, it is difficult. But I think that we are still at least a couple of decades from having self-driving cars that can do everything and everything is possible. So, I guess the question is how much more will we go through in these 50 years to have cars that can drive the environment and the environment needs to be better for us, especially for our children. That is the question.”
I mean, that’s not bad really. Perhaps this AI thing actually works!
What is AI?
In computer science, artificial intelligence (AI), is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans. Generally, the term ‘artificial intelligence’ is used to describe machines/computers that mimic things that humans associate with other human minds, such as learning and problem solving.
Some prefer the term assistive intelligence (AI), because they suggest that machines are not intelligent, they just look like they are. There are four types of artificial intelligence but the two we are interested in for now are:
- Reactive machines – The most basic types of AI systems are purely reactive, and cannot form memories or use past experiences to inform current decisions. This type of rule-based intelligence involves the computer perceiving the world directly and acting on what it sees. It doesn’t rely on an internal concept of the world. These machines behave exactly the same way every time they encounter the same situation
- Limited memory – This classification contains machines that can look into the past. Self-driving cars do some of this already. For example, they observe other cars’ speed and direction, which requires identifying specific objects and monitoring them over time. These observations are added to the self-driving cars’ pre-programmed representations of the world. These pieces of information about the past are transient so are not saved as part of the car’s library of experience.
Bottoms up
There are advantages and disadvantages for both top-down and bottom-up AI systems.
Top-down (symbolic) approach:
- Hierarchically organised (top-down) architecture
- All the necessary knowledge is pre-programmed in the knowledge base.
- Analysis involves creating, manipulating and linking symbols
- The program performs better at relatively high-level tasks such as language processing.
Bottom-up approach (neural networks for example)
- Models are built from simple components connected in a network
- Relatively simple abstract program consisting of learning cycles
- Program builds its own knowledge base and common sense assertions
- Intelligence emerges from the interactions of large numbers of simple processing units
- Built-in learning mechanism, results in adaptivity and flexibility
- Better able to model lower-level human functions, such as image recognition.
Rule based software can be a valuable part of a driverless car’s toolkit for high-level control applications like route planning, and to manage low level activities, such as checking the status of the fuel tank. However, rule based artificial intelligence tends to break down in unstructured environments, leading some roboticists to refer to top-down AI software as ‘brittle’.
This figure shows a representation of methodologies and how or where they can be used.
Deep learning
Deep learning is a class of machine learning algorithms that use multiple layers to progressively extract higher level features from a raw input. For example, in image processing, lower layers may identify edges, while higher layer may identify human-meaningful items such as a car, a person or an object.
Multi-layer deeply connected networks perform non-linear transformations on incoming data. It is often modelled on biological systems, such as deep neural networks or convolutional neural networks. Systems are ‘trained’ by applying a large, labelled training set in either a supervised or unsupervised optimisation process:
- Supervised machine learning is where the program is trained on a pre-defined set of examples, which then enable it to reach an accurate conclusion when given new data
- Unsupervised machine learning is where the program is given a large quantity of data and must find any patterns and relationships that it contains.
Summary
Whether we think of AI as artificial intelligence or assisted intelligence, it is a fascinating and complex subject. It is an important part of the automated driving vehicle that may ultimately lead to them becoming autonomous. There are two main types of AI and perhaps for an easy distinction we can describe these as one that simply follows rules and one that learns every time it encounters a new situation.
The technologies in general associated with automated driving vehicles (ADVs) are really interesting but it is also important to consider how these will affect the repair and service industry. Will a car make its own booking for a service when an on board monitoring systems tells it that something is out of adjustment or it is near the time for an oil change?
The final article in this series will look at some of the consequences for our trade.