The world’s leading technology companies regularly publish research papers in the top AI conferences as part of their strategy to build global competitive advantage. These companies include AI behemoths such as OpenAI and Anthropic, heavy users of AI in the tech stack such as Amazon and Meta, and early stage tech companies such as Pluralis and Marqo.
It is very unusual for Australian companies to do the same, but this needs to change.
Not publishing may have been viable in the past, when our resources made up for our lack of innovation, but AI is eliminating the barriers that once protected Australian industry from international competition.
Uber is a shot across the bow because they showed that they can run our taxi industry from the US. The competition that is coming for Australian companies is digital- and AI-first, has a much lower cost base, much smarter technology and an ability to recruit Australian customers directly from servers in Seattle, or Beijing.
Having a corporate policy that encourages workers to undertake research in AI and publish in the best AI conferences enables high-tech companies to attract the best talent in AI. The best talent wants to work on the best problems with the best people in the best companies.
It is staggering to see workers turn down $100 million sign-up bonuses to stay in companies that give them the best development opportunities. Those companies that can attract and retain high quality talent have the best chance of outthinking their competition, and surviving.
University of Adelaide Professor Anton van den Hengel
For individual workers, publishing in the top AI conferences represents concrete evidence that they can compete at the highest level of AI innovation globally.
The financial incentives for being classed as ‘top talent’ are enormous, so potential applicants are very motivated to demonstrate that they are at the cutting edge. In order to have a paper accepted at one of the top AI conferences you need to have an original idea that advances the current state of the art in AI enough that at least four experts think it is worth publishing. You also typically have to implement the current state of the art so you can compare your new contribution against it.
Irrespective of the actual content of your paper, the fact that it has been accepted is a good indication that you understand and are working at the cutting edge. If you’ve had a series of AI papers accepted over a period of years it demonstrates a commitment to staying at the cutting edge of a very dynamic field.
The best talent are unlikely to work for a company that doesn’t allow them to publish because it is career suicide. They need publications to get their next job, so not publishing is not an option. One of the reasons companies thus have to publish visibly is that it is essential if they want to attract these people.
The best reason for a company not to publish is that they don’t need cutting-edge AI expertise. It is only businesses that are aiming to use AI to build new products, outcompete incoming competitors, or build global markets that need cutting edge AI skills. It is very difficult to identify the new products and opportunities that AI makes possible without a deep understanding of what the technology can do, and how to get it to do it.
AI is also an area where deep technical skills matter, because small performance improvements can make all the difference. Google built their search business on having AI that is slightly better than the competition, because that is enough to get most of the traffic.
The worst reason for a company not publishing is that they think that time staff spend publishing is time they could spend working on products. This means they misunderstand the value proposition.
The time staff spend working on research and publications is time spent developing a deep understanding of the technology, and an ability to build new and unique AI. Staff have to understand the latest research to be able publish, and doing so means they will make better decisions about what will work best for their employer, and how to make it work quickly. Coincidentally, it also help keep staff engaged with their jobs because it’s interesting, and because it reinforces the idea that their employer cares about their careers.
Estimates for the percentage of AI projects that fail are as high as 95 per cent. The main reason for this failure rate is that the projects are ill conceived or badly implemented, due to a lack of expertise. In the short term it thus looks like staff engaged in publishing are fractionally less productive, but in the long term it means they are far, far more effective.
One of the most insidious of Australian narratives is that Australian companies can’t compete at high-tech, and should thus stick to applications of other nations’ technologies. One implication of this narrative is that there is no need to publish if you’re technology strategy is to be an also-ran.
The narrative is obviously false, as there are great Australian companies competing at the cutting edge, and many outstanding Australians who are working for the leading high-tech companies globally.
The talent pool isn’t just current and previous Australians of course, there are a host of highly qualified potentially future Australians working overseas who’d love to work here. Australia also has world class AI research, and seems to be able to compete in fields like quantum computing, so the self-defeating narrative is hard to fathom.
The boards of Australian companies that can see that international technology companies are coming for their market share should make sure that their companies have a strategy to build global competitive advantage through their tech stack. This means attracting the best talent, and this means a commitment to carry out and publish world class research in AI.
Do you know more? Contact James Riley via Email.