On 26.09.2022, artificial intelligence was again the topic of the Millásreggeli radio show. Levente Szabados, co-founder and senior advisor of Neuron Solutions, gave us another very interesting conversation.
“Those who don’t understand the past don’t understand the present and not even a little bit the future.” With this statement Levente takes us back to the past to answer the hosts’ question “Who owns AI?”. Originally, when AI started to develop, it was a partly military, state-funded technology (’50s onwards). However, after the initial AI developments and projects failed to live up to the exaggerated expectations and promises made of them, public funding dried up. The next breakthrough came in 2012, when funding returned in the form of corporate and open funding, which, along with the boom in open science, also boosted the field of AI. Today, it is standard practice to publish a new result, usually with code and process, when it is published. And this is indeed the case. On almost any topic, there is a wealth of publicly available code, take Deep fake technology for example, where around 1300 code repositories are publicly available.
But how is this possible? “How is it that technologies developed with huge investment of time, money and energy are just shared for free?” – was a very legitimate question from the presenters.
Of course, it is not uncommon for only a dumbed-down version to be published, but surprisingly it is worth it to publish the whole piece. Collaborative development of publicly available code, known as Open Source, is one of the most effective development paradigms today. The secret to its success is simple: with multiple eyes, any bugs are pretty easy to spot.
Moreover, in many cases, the competitive advantage is not lost by making the algorithm public, because the code alone is not enough, it is of little value without the right data. Take Facebook data, for example, to which we will never have access.
So who has the strongest AI application? Good question. But we need a metric. What does strongest mean in this case? Does it mean the bigger the data, the bigger the models they can create? Because in this regard, the “big tech” companies are leading the way, think of Facebook, Google, or Open AI, for example. So in terms of size and capacity, they lead the ranking.
However, we can consider the specialisation of the models, and then others are ahead. So it is difficult to define this power and force issue well.
For example, the ability to filter data, to use it well, is important and crucial, otherwise one can easily find oneself confronted with the phenomenon of “garbage in, garbage out“. A very strong industry has been created for annotating images to ensure data quality. Or a particularly strong strategy, with little energy investment, to transform existing processes to produce quality data for us as a “by-product”. For example, in a manufacturing process, automatically recording data on damaged or intact products, possibly by taking a photograph, and thus easily (avoiding repetitive labelling work) obtaining high quality annotated data.
But there is one more component to the formula, because neither data nor prediction is the bottleneck, but there is the question that overrides technology: what should be predicted and what should not?
So who has the power? That may be difficult to say, but in terms of AI, Levente has tried to define for us what success is. It has been said that while the tools are available to many, few people know the nature of these tools well enough to know how and what they could be used for. Not forgetting, of course, what it is allowed to be used for – Levente mentions the example of race, ethnicity or the identification of one’s person. Indeed, it raises ethical questions if our online behaviour is used to determine our ethnicity or identity through AI and then used to make business decisions about us.
Don’t worry, this is not the end! The conversations with Levente will continue, follow us and don’t miss the next episodes!
You can listen to the full conversation in Hungarian in the following Youtube video: