Artificial intelligence: Is everything going faster and faster now?
Demis Hassabis is one of today's most influential neuroscientists and computer scientists. He founded DeepMind, an Artificial Intelligence research company, in 2010. He recalls, "In 2010, nobody was talking about AI. We could barely scrape together two cents for it in the investment world."
But since then, mathematical computing has advanced rapidly. One incredible recent example: at the Center for Computational Quantum Physics in New York, AI has been used to trim a quantum physics problem from around 100,000 equations to just 4. The punch line: without any reduction in accuracy.
What is AI?
The term "Artificial Intelligence" was born in 1955 at a workshop in Dartmouth (USA). At that time, it was still a matter of programming basic logic rules for computers. In the early 1970s, this gave rise to the first "intelligent" recommendation systems. As such, AI is still a generic term for various programming directions. Machine learning is a subset of this term.
This clarifies a key point: Strictly speaking, artificial intelligence and machine learning are not identical. However, for the past 10 years or so, most of the groundbreaking advances have taken place with one of the many Machine Learning approaches.
We speak of machine learning when automatic optimization processes are achieved via algorithmic feedback loops. In this way, computer systems learn amazing skills without being precisely programmed by humans in advance.
How exactly is AI defined?
Interestingly, the term is not particularly well defined, rather it is simply a developmental goal for advanced computer science. We can refer to SAP here: "The general idea is to converge machines and significant functions of the human brain: learning, reasoning, and problem solving."
So, the "general idea" is an approximation, specifically to a model, in this case human intelligence. This is not to be taken literally: AI has long since achieved superhuman performance in terms of precision, speed and in recognizing hidden correlations in entire mountains of structured data.
How is AI being used today? Everywhere.
AI is now found in hospitals, cars, homes, all sorts of everyday devices, large machines and factories, administrative and research departments. It can automatically translate 50 languages into any other, analyze X-rays, develop optimal designs, control sprinkler systems, reduce energy waste, predict demand, increase sales, customize web pages in real time, predict protein folding of molecules, make cars drive themselves down the highway, and most recently, generate artificial photos from verbal input.
We see AI as the most fundamental innovation of our time because it is now accelerating the pace of innovation itselfworldwide, for example in generating ideas for new medicines, chemical compounds or revolutionary building materials. What was an unrealistic niche area of computer science 15 years ago is now commercial mainstream.
What are the main application areas of AI?
If we go by invested capital, here are the top areas, according to Statista (2019):
Since 2019, the order of the lower items have shifted here and there, as investments have more than doubled since then. But you see: Machine Learning is in practice today in almost everything called Artificial Intelligence. Somewhat obscure is the area of "ML applications," into which most investments flow.
Behind this are integrated AI modules that enable certain features and functions at any point in a piece of software. The route function in Google Maps, spam filters in an email account, or the prediction for the arrival time of an Uber driver are three examples of hundreds of thousands. What we have been doing at avvaneo since 2019 would also fall under this heading.
What does avvaneo do with AI?
Our "purpose" is optimization. We optimize financial processes with automation software, then optimize our automation, and finally we are constantly looking at how we can improve our optimization strategies. Logically, we have been working with AI since 2019.
The first result was avvaneo Expedite, launched in 2020, an AI solution that increases the level of automation of your financial processes beyond the previous limits with rule-based automation. With Expedite's AI, precise process recommendations or allocations can be generated from your existing invoice data and implemented automatically.
How does AI work in practice at avvaneo?
Since launch, we have implemented many different use cases. Our AI module has already frequently proven helpful for coding incoming invoices or purchase orders, discovering the correct approval path, detecting anomalies ("anomaly detection") automatically notices anomalies in all processes and reports indications of possible fraud.
Our experience with individual use cases continues to grow, which benefits all our customers. Recently, we added the matching of incoming invoices with purchase orders as well as a duplicate check to automatically and proactively avoid double payments.
How much more efficiency is achievable with our AI solutions?
The Dachser Group from Kempten (Allgäu) is one of Germany's largest logistics companies, with around 30,000 employees worldwide and annual net sales of almost EUR 6 billion. For Dachser, we have further automated accounts payable processes and purchase requisitions. With the help of avvaneo Expedite's integrated AI. In other cases, much higher levels of automation are possible. Of course, this always depends on the specific requirements. But that's the point: the more challenging the individual requirements at a company are, the greater the efficiency gains that can be achieved with "self-learning" services, i.e., with artificial intelligence.
We will continue in Part II.
In Part II, we go deeper into the details. First, we describe our applications in more detail. Second, we look under the hood and demonstrate how the functioning of AI can be better imagined from the user's perspective.