What is artificial intelligence capable of when it comes across complex human science data?

The presentations given at the HAiLife webinar confirmed the centrality of AI development in the utilisation of human sciences data.

Lately, artificial intelligence has frequently been on the lips of those interested in technology. In particular, examples of the evolving capabilities of chatbots to assist humans in a conversational manner in tasks such as information retrieval, text creation and even coding have been topics of considerable current interest. And indeed, the topic of the second seminar of the HAiLife project was the business opportunities provided jointly by human science data and artificial intelligence. What is artificial intelligence capable of when it comes across complex human science data? The webinar featured two interesting presentations, the first of which showcased HeadAI's AI-based business that is already underway, and the second a product of Loihde AI, which is still under development.

HeadAI's artificial intelligence makes expertise visible

The first presentation was given by Harri Ketamo, who has years of experience in cognitive sciences and artificial intelligence in, among other things, game development. Ketamo has also extensively published research on the subject. Currently, the company he established, HeadAI, is developing semantic AI for visible decision-making.

At first, Ketamo briefly described how HeadAI has been developing its own artificial intelligence and algorithms for years, and that in practice, the computation takes place on their own closed servers. One of the advantages in this is that it also allows using various sensitive materials in a secure manner. In addition, compared to open systems, this setting enables more precise control of the AI models with regard to e.g. the various biases built in the data.

At present, a large part of the important decisions made by individuals, companies and society concerns competence. Ketamo presented several examples of how competence data can be utilised in support of decision-making. Data not only depicts individual competences, such as mathematical competence. Above all, it makes it possible to describe how different competencies are interrelated. In practice, HeadAI's analysis enables making the competences acquired from different contexts visible and consequently also the strategic further development of such competences. For example, we may find that by supplementing a certain area of our own competence profile, we can create a whole new kind of value with a more diverse combination of skills.

On a larger scale, the situational awareness picture of competences enables better decisions related to education and competence. From text data, for example, it is possible to extract a picture of the areas on which demand for expertise will focus next and decide on the measures to be taken in response to this. At the same time, text-processing AI is already able to generate courses on different content areas almost autonomously, and this is what HeadAI has been implementing – even before the most recent discussion sparked by chatbots. However, the interpretation of data and results must remain critical. For example, different keywords describing competence do not necessarily take into account the overlapping competence under another keyword (cf. statistics and artificial intelligence). Therefore, human critical thinking in the 'loop' is still needed. Time will tell what the rapidly evolving analysis of textual materials will enable next in terms of business.

TurvAIsa by Loihde is the support intelligence of the future

The second presentation of the webinar was given by Loihde AI's data scientists Laura Laaksonen and Seppo Nyrkkö. A Lead Data Scientist at Loihde AI, Laaksonen has 20 years of experience in the development of audio and sensor technologies and in the application of speech processing and machine learning in particular. Nyrkkö studies the machine processing of linguistic expressions with the help of artificial intelligence and structural models. In their presentation, Laaksonen and Nyrkkö introduced the audience to the theme of speech emotions in the context of security. Loihde's TurvAIsa, currently under development, addresses a practical need: In nursing, workers regularly face the threat of violence.

Among other things, a system is being planned that would be able to recognise aggression from speech emotions in challenging and threatening patient situations. This could be used to prevent such situations from occurring and, if necessary, to call help. The application is based on decades of research, in which emotions have been found to manifest themselves in speech, not only in words, but also in various non-verbal cues that can be recognised from the sound signal using various techniques. Speech emotions are currently under intense research, and different ways of utilising speech signals have developed by leaps and bounds. At the same time, plenty of challenges remain related to the context of use, for example. For example, factors related to environmental acoustics and background noise require technology to be capable of identifying and filtering essential information for use. Loihde's data scientists are currently developing solutions to address these factors.

In practice, the future application could work in the background of nursing without any major changes to the employees' ways of working. The threshold for setting off an alarm and a call for help can be adjusted where necessary, and there may be other ways to trigger an alarm in addition to a speech emotion. The aim is to carry out practical calculations in terminal devices as far as possible, which has the advantage of, among other things, lower power consumption and more stable operation of the entire system. In addition, the system does not automatically store personal data.

In the concluding discussion at the end of the presentation, several other applications for similar technology were raised. Emotions are part of interaction in many different situations and thus a potential channel for supporting interaction in workplaces across the board.

Human-machine dialogue continues – cooperation between human sciences and data sciences is needed

The presentations given at the HAiLife webinar confirmed that AI development is a central factor in the utilisation of human sciences data. Human understanding is often not possible with simple algorithms, but instead rather complex calculations are needed in the background. On the other hand, the development work is based on long-term research that in the case of emotions, for example, stems from theories developed in the human sciences. Therefore, it could be assumed that future applications of artificial intelligence will also increasingly require collaboration between human sciences research and data sciences. The HAiLife project addresses this challenge on its part. Join us for upcoming seminars and stay in the conversation!

Last updated: 15.11.2023