How Artificial Intelligence changes research

Some background 

Technology in the artificial intelligence field has developed at a rapid pace in the last few years. While AI as an idea and theories about it have been around for a long time and in practice it has existed in a more limited forms for many years. The recent boost given by both advancements in chip technology and NVIDIA enabling developers accessing their graphics cards GPU cores advanced parallel processing capabilities through their CUDA API interface and then the past years publicity around large language models (LLM)s like ChatGPT have been pushing things into the limelight and public awareness. Encouraging researchers and developers to go further than what has been possible before with the AI technology. 

While generative AI in the form of LLMs and AI generated graphics are the most visible forms of the technology commonly available to people, there are near infinite amount of possibilities how the AI technology could be used overall. Basically, anything that can be represented in a digital form can be assessed and often a pattern emerges that can be identified, analyzed, filtered and replicated. Autonomous learning and decision making are also major fields of development, without which truly self-driving vehicles would be an impossibility. While the level of technology currently available is still in its infancy compared to where we will be in a few more years. The improvements in speed, quality of results and reductions in the resources needed from the platform to run these models have been significant during just the last spring alone.

AI tools and functionality is now popping up everywhere, where just a year ago one would have been hard-pressed to convince a company why they should be using the technology, let alone which one, how or where. The steep learning curve will cause some growing pains as always with new technology and not all AI services are equally good or useful, but as the AI field settles in as the new norm, if not a fixed form, then at least a direction emerges. And that is towards more multiuse platforms that run faster, more secure and on lighter hardware requirements than ever before. As the complexity of the AI models and what they are able to do increases, this makes it both imperative and also more difficult to optimize them and ensure the reliability of the results. These are just among the few problems that need to be solved in the coming years, so the research community has its work cut out for it in the years ahead.

 As our lives are being saturated by the omnipresence of AI technology, research is no exception. The way we do research in the coming years will be fundamentally different than just a few years ago. No longer can it be presumed that all of a publication has been written by a human being, or that all of the interview data has been sourced from live individuals, rather than having been generated based on a pattern previously recorded from older interviews. 

New tools bring new challenges like repeatability, human in the loop, ethics in the amount of AI tools used and how its use is being disclosed in the publication. But also opportunities, like sorting through larger datasets than previously possible and recognizing patterns too big or complex for human to recognize with conventional tools. Generative AI particularly challenges us to be preceptive and to think more holistically. Where do the limits of human ability lie and where the strengths? What is worth automating and what is worth preserving as a human-to-human interaction or human-to-machine interaction? More novel is the networking of machine-to-machine AI tasks, with little if any human involvement in between the original task giving and analyzing the results. While this can save time significantly, possible error propagation gets harder to detect and prevent. This is still the inevitable direction in which things are headed, not just for time savings but in order to improve current methods with feedback loops and enable more complex chained task, closer to human performed refinement with trial, error and self-review of the results and combining the use of different tools or modules in one task. AI is also becoming cheaper, faster, more mobile and less dependent on an internet connection. Combining smaller but nearly as good AI models with filters and hardware designed to run them, local mobile AI is coming to change the way we view and use AI tools in the immediate future.

In the end AI models are still tools, albeit very advanced ones at that. By their very nature their function and use is different from what we have been used to. Able to interpret pictures, written and spoken language, read documents, respond like a person vocally, with real sounding language and rational sentences even about complex topics. This requires a new mindset from us, in order to express our unique voices and abilities trough these new improved tools and still retain focus what is relevant. The age of AI has just begun and it’s not going anywhere, so it is up to us each individually and collectively to shape the future into a form where these AI tools serve us to everyone’s advantage. There is no better time than now to get involved, while things are still in flux.

Author Kimmo Kauria