After the outbreak of Corona, it was found that artificial intelligence had predicted the pandemic based on data. Does this show us that we should rely more on data-based decisions in the future? A definite yes and no. People have always been dependent on information. Thousands of years ago, people needed to know which plants were edible and which were not. They have passed this knowledge down through generations. A decision to eat the wrong plant could be fatal in case of doubt – in the best-case scenario, it just caused a stomach ache.
We always make decisions based on available information – consciously and unconsciously. Processing information or data is, therefore, nothing new. So, when making decisions about computers, we should ask just as critically what data is underlying and what information was used to train the AI? The “new” addition is the digital processing of information. Information processing works at the speed of light, and a machine can consider infinitely more information than humans.
So the machine depends on the human performance of information processing?
A person’s “memory” is limited. He can juggle a maximum of five to seven factors in a conscious decision – with a gut decision. It is up to 20. We can do about 1013 analog arithmetic operations per second. A machine can already do a lot more today. The supercomputer BlueGene / L from IBM, for example, manages up to 3.6 1014 floating-point operations / second with double precision.
The machine, therefore, has the speed advantage, including the amount of data it can process. However, it also has two significant disadvantages:
- In contrast to the human brain, it cannot collect any information by itself. We have to “feed” the machines and train them (deep learning).
- The human brain is highly networked, resulting in massively parallel processing that computers cannot yet provide. However, chip designers are working to overcome this drawback.
The robo adviser of the company Minto AG, for example, landed an absolute stroke of luck in the corona crisis in the area of shares in real-time trading. Based on AI, market risks were identified in good time, and stocks were sold within seconds. As you can see, technology speeds everything up and does a lot more in less time. If you are fast and have enough data, you can make more profits – so the simplified formula.
So AI Decisions Can Help Us Move Forward?
Definitely, but we have to keep in mind that decisions can be just as wrong – as they are with humans. Large amounts of correct data are the decisive factor in AI. Distortion due to insufficient data can lead to bad decisions, known in technical terms as the “bias effect.” The assumption that computers and AI are infallible is wrong because we “feed” or program them. As much as decisions in real-time – without human intervention – are beneficial, they are ethically tricky in many places.
This applies not only to obvious cases in the military, where life or death is at stake but also in securities trading. If unforeseeable damage can occur, a decision by an AI should not lead directly to action but rather be confirmed by a human. This also includes moral principles, especially when a machine does not wholly picture the world. So I see AI primarily as an assistance system that suggests decisions to us. In the long term, however, we will not be able to avoid teaching the rules of the machine for morality.
So Do We Need Ethical Principles For Machines?
I’m a fan of applying guidelines that were developed in the real world to the digital world. For example, who is liable for a wrong decision? A control mechanism is essential. We have a decisive advantage with machines, which I would like to point out as an opportunity: Since there is a complete data database, decisions can be traced transparently. So why was this done, or what data was “fed” with.
This is not possible with a person. It would contradict the principle of the protection of the individual. Data for the development of AI must therefore be carefully selected. The quality and thus the degree of freedom from errors of an AI is determined by the data selection -the better the data, the better the decisions. As with the car, it is also not allowed to drive on public roads if it is not regularly technically checked.
What Are Good Data-Based Decisions?
It isn’t easy to give a general answer. A “good” decision should always bring added value. It should move things forward. From a company perspective, you could also say that the best decisions represent the best sustainable cost-benefit ratio. Data can help here since if a good database is available, patterns from the past can be recognized, which has led to an excellent cost-benefit ratio. The task for an AI is the same as for a human being to look for these patterns.
What Is Your Prognosis For The Future?
Making faster and better decisions with the help of data and AI technologies are coming.
The advantages are apparent. If we invent new products based on customer usage data, this offers enormous potential for added value. Or, if we repair devices in good time based on data, valuable resources are saved. But I think that the question of the quality of software-based decisions will continue to move into focus. They can offer the speed, we can already see that today, but the holistic quality of these decisions will be the subject of developments in the next few years.