2024 in AI.

I wish everyone a happy, healthy, peaceful, and prosperous 2024!

I will write and share a lot of information and perspectives about Artificial Intelligence and Robotics in the upcoming twelve months while working on the living Appendix for "Romy & Roby and the Secrets of Sleep." Here are my few cents for today.

ChatGPT 3 started the movement of democratizing AI, showing the possibilities of Machine Learning to the masses. While journalists and bloggers from different backgrounds and with varying knowledge about AI were busy hyping Large Linguistic Models, AI ethical risks, and leadership drama at OpenAI, few important truths were left out of sight.

1) Transformer architectures are costly to develop, establish, and maintain:

So far, only very few companies can produce ChatGPT cousins at sensible costs. Nvidia won the 2022 Generative AI race because it implements the hardware to train LLMs and similar architectures.

Mid-sized AI businesses spend 25 percent of their revenues on the cloud and a minimum of 15 percent on cleaning and preprocessing data. I am sure there will be more innovation to ensure training data can be prepared more efficiently. Still, without changes in architecture behind Deep Learning and transformer-based solutions,? new AI companies will fight against challenging cost dilemmas.

In the first book of the "Romy&Roby" series, one protagonist is reflecting on whether the family needs to build its own data center under the pond in the garden to address mounting energy bills coming from the presence of an Android in the house. Energy costs are essential in genuinely democratizing any technology, and we are far from a scalable and reliable solution.

2) AI businesses might go the way of Big Pharma due to the cost dilemma:

European AI Act added an additional cost burden to AI innovators. There is no wonder Big Tech spent hundreds of millions on lobbying efforts in Brussels. I think the nascent AI industry might go the way of Pharma, only that Pharma had over a century to come to what we know today. It will be much more challenging to build a sizable Machine Learning business. Small competitors will suffocate in costs and be snapped from the market by twelve to fourteen American and Chinese giants.

3) GPT-type technologies and LLMs won't bring us AGI:

GPT-type models won't bring us Artificial General Intelligence. There aren't ways to remove hallucinations from LLMs because the distribution shift of these models can't be solved. These applications will be helpful for many industries and in many use cases, but no more than this.

4) Copyright questions aren't only to finance law firms' business models:

Copyright issues will surface more and more. No wonder NYT&Co went after Open AI. Non-profit businesses, museums, and libraries will be faced with whether they should legally challenge AI makers to use their data for model training and development. There won't be uniform answers. Interestingly enough, AI forces us to think about what our businesses are about, how we want to scale, and what value we want to put in front of our customers, partners, and communities.

5) Synthetics: Quo Vadis?

Generative AI will further flood the world with synthetic data. The industry and regulators will need to act proactively on detecting synthetics. Statisticians will discover behavioral patterns in which these synthetics differ from real-world data. In many cases, these data are helpful to develop a new drug, finalize the composition of a song, decrease costs in designing a movie set, or mimic population data in case actual data can't be used due to privacy protection. I hope for a hysteria-free dialogue on what we do with digital assets in businesses if they are augmented with synthetic data, how to evaluate these, and how to avoid their unnecessary or overextensive use for better product and process design.

6) Small is beautiful:

I hope we will see better innovation in small data. This will allow traditional businesses and small AI innovators to build sustainable businesses at a healthy cost base.

7) Cybercriminals will further endanger everything around us while using Machine Learning:

Cybercriminals use AI and Generative AI as weapons to defraud businesses and consumers, steal personal information and identities, cause business interruptions, and destroy the reputations of enterprises and individuals. Here, I call on regulators to produce more severe rules to manage Digital assets. Exposure quantification, security orchestration and incident management automation, and capabilities to spot and remove bots from social media accounts must be on the agenda of any leadership team in the same way as weekly cash flow analyses.

Looking forward to talking to people interested in AI and willing to challenge the status quo of general technology literacy in Europe and around the world.

?