Trends

Data & AI: Is weak AI gradually becoming stronger?

The current “Tech Trend Radar” from ERGO and Munich Re covers 44 trends in five topic areas, which we take a closer look at here on //next in a series. We started with our detailed analyses on “Wellbeing” and “Hyperconnectivity”, and now we turn to “Data & AI”: What is this trend field all about – and which individual trends are hidden behind it?

Tech Trend Radar, Themenbereich Hyperconnectivity

The “Tech Trends Radar 2022” (TTR2022) defines the trend field “Data & AI” as the "home for data-driven technologies and business solutions based on artificial intelligence" (AI). Key applications in this field include, for example

  • “Computer Vision” (machines learn to interpret images) as well as
  • “Natural Language Processing” (NLP), with the help of which machines understand human language and use it themselves – for example in the form of PhoneBots, which are also used at ERGO.
  • “Machine Learning”, on the other hand, enables independent decision-making, while the TTR2022 makers refer to
  • “Reinforcement Learning” and “Knowledge Graphs” are important principles that help machines to become artificially intelligent.


AI is making enormous progress ...

But what has been groundbreaking in this complex field in 2022? First and foremost, “multimodal AI” – a big step forward in making AI-calculated results more and more accurate. For Daniel Grothues from ERGO, one of the two project leaders behind this year's Radar, “AI democratisation” is one of the most significant milestones in recent AI history: “Basically, this is a new term for 'explainable AI' – i.e. the possibility of making AI altogether more accessible and much more widely usable (see below).” You can read a detailed interview with the two project leaders of TTR2022 on its origins as well as on what they see as the most important trends here.

... but still far from real intelligence!


Despite all the recent and ongoing progress in the field of “Data & AI”, it is important for the two experts to make it clear: It will be a long time before artificial intelligence – in this case, according to the TTR2022 definition, it would be “artificial general intelligence” (AGI, see below) – is able to compete with humans on a broad scale. We are talking decades here, if it ever gets that far. This is also the assessment of Mark Klein here on //next: even a four-year-old child is still far superior to even the most artificial artificial intelligence, the ERGO CDO was quoted back in 2020. Which is why the machine intelligence that prevails today refers to a “weak” AI that has been trained for narrowly defined tasks. In contrast to a strong AI, which actually imitates human intelligence. The latter is still dreams of the future – quite clearly!


Which trend has which maturity level?

Details on the twelve trends in the trend field “Data & AI” can be found below, but first let's remember: The study divides all trends into the four different maturity levels:

  • hold (“Put on the watch list”)
  • assess (“think about what this could mean for your company”)
  • trial (“first initiatives should be launched in the most affected business areas”)
  • adopt (“take full advantage of this technology!”)

But which recommendation applies to what?

Five times “adopt”…

According to the “Tech Trend Radar 2022”, insurers should implement the following five trends:

  1. Computer Vision: As outlined at the beginning, this trend encompasses the most diverse methods for capturing, processing, analysing and interpreting digital images. It is therefore considered one of the most important drivers of applications that require visual data from sensors – be it robots, drones or devices in the Internet of Things (IoT). A classic example is AI-controlled vehicles that navigate through their environment by interpreting the images from their image sensors ever better. But industries such as retail also benefit by automatically making product suggestions to their customers while they are shopping – based on images of items they have already put in their shopping trolley.
  2. Digital Twin: Twins resemble each other like one egg to the other. This also applies to so-called digital twins: virtual images of real objects. We have already described on //next why these representatives are so valuable in the industrial and manufacturing sectors or in urban planning and development: They make planning, testing and optimisation so much faster, more convenient and cheaper. According to the “Tech Trend Radar 2022”, the number of these digital twins will reach “hundreds of millions” in the medium term – a market penetration that should also be highly interesting for the insurance industry through cooperations and partnerships: "adopt!

  3. Machine-driven Decisions: The volume of data worldwide is increasing rapidly, and algorithms have long been able to compile, sift through and evaluate all the information on specific issues. The machines are also succeeding more and more reliably in the next logical step: drawing conclusions and making decisions. Depending, of course, on how much data is available and in what quality. The insurance industry could use this trend above all to have machines prepare decisions – and then have humans make them. The keyword here is “augmented decision-making”.
  4. Natural-language processing (NLP): The processing of human language by machines facilitates interaction between humans and computers and significantly increases efficiency in ERGO customer service, for example. In the meantime, the technology has advanced to such an extent that even experts find it difficult to distinguish its output from human content, even if the algorithms still lack semantic and contextual polish. Nevertheless, the TTR2022 advises insurers to exploit the potential of NLP in as many applications as possible along the value chain: from communication, document analysis, matching and classification to chatbot enhancements and translations.
  5. Digital Assistants: These virtual assistants – the most prominent representatives are probably Alexa, Siri and Google Assistant – are not only driving the shift from text-based to voice-based communication between humans and machines. They can also increasingly offer advice on their own and anticipate our actions and requirements. And: they are getting better and better the longer and more often they assist us. According to “TTR2022”, they will become indispensable in everyday situations and in the care sector in the future – and they also hold great potential for the semi-automation of complex business processes such as claims management in the banking and insurance sector with its gigantic treasure trove of data.


... three times “trial”…

  1. AI Democratization:  Behind this trend are standards, approaches and applications that make this highly complex field of technology accessible to a broader range of developers. AutoML, for example, shortens the process of machine learning in such a way that programming-intensive intermediate steps are no longer necessary and basically only the data need to be fed in. Low-code and no-code platforms, on the other hand, enable even people with little programming skills to develop applications – via point-and-click, via graphical interfaces or based on model-driven logic. Since such democratisation steps significantly shorten the time for developing and testing new AI applications, insurance companies should, according to TTR2022: “trial!”
  2. Knowledge Graphs: We humans interpret information continuously and can sort and evaluate it by matching it with our background knowledge – and what we call “common sense”. This trend includes a new wave of AI applications that combine and reference data from different sources across silos in a way that gives it context and processes and stores it in a similar way to the human brain. The result is a complex information network from which missing facts can be derived much better than from a conventional database. Prominent applications are improved search engines, automatic question answering services and social networks. But basically, every company that is confronted with disconnected, heterogeneous and constantly growing amounts of data benefits from this trend: it channels the flood of data, facilitates queries and helps integrate company and external data, leading to new insights and efficiency gains.
  3. Reinforcement Learning (RL): In this iterative learning process, an AI model is confronted with game-like situations and attempts to arrive at a solution by means of trial and error. Areas of application include self-driving cars, industrial automation, retail, finance and healthcare. Although considerable progress has already been made in this trend area, RL is still considered a research area. First use cases in the insurance industry can be found in management and underwriting.


... and two times “assess” ...

  1. Quantum Computing: This trend promises unprecedented computing power and could trigger a new wave of technological developments in the next five to ten years. We have already outlined here how exactly quantum computing works. In 2021, ten leading German companies, including Munich Re, established the Quantum Technology & Application Consortium (QUTAC) to unlock the enormous potential of these trends for the insurance industry and beyond. The aim of the consortium is to identify, develop, test and exchange applications, thus supporting the transition from basic research to use: “assess!” 
  2. Multimodal AI: This is what the TTR2022 means by various AI models that combine inputs from different data sources to achieve increasingly accurate results. These data sources can vary – be it text, audio, images or other formats. By analysing such different types, AI can contextualise the data to perform better than traditional unimodal AI models. Typical applications here are also transport (advanced driver assistance systems and human-machine assistants) and security or payment systems. Healthcare is also exploring this area – for example, to integrate medical images into electronic health records. 

… und one time “hold”

  1. Artificial General Intelligence (AGI): As already touched on above, this trend is only a topic for science fiction enthusiasts and “what if?” planning games for the foreseeable future. Because today's AI technology – and that of tomorrow – lacks common sense, real intelligence and methods of self-preservation or reproduction. All progress in the field of AI – impressive as it may be – is limited to “weak AI” – i.e. special applications for limited purposes. But should this change, one of the future “Tech Trend Radars” will identify it early ;-)
     

In the next article in this series ...

... let's take a look at the trend field “Cyber & Crypto”: Which ten trends are hidden behind it – and which applications are already in use?

The German version of this article is here: Data & AI: So wird schwache KI immer stärker 

Most popular