Various AI-based services on Generative AI1 have made a big impression in recent months. While these tools are impressive in many ways, they have also led to concerns about the risks that advanced AI can pose. Elon Musk, Steve Wozniak and other well-known figures are calling for the development of artificial intelligence to be temporarily halted. These 1,100 individuals have published an open letter calling for a temporary halt to AI solutions that are more advanced than GPT-4. The letter emphasises planning and management. It argues that powerful AI should only be developed when its effects are positive and the risks are managed.
Italy temporarily banned ChatGPT because they believed the technology could lead to violations of the GDPR, but the ban was lifted after OpenAI explained and possibly implemented changes to address Italy’s concerns. However, countries that block or severely restrict the use of this new technology are likely to fall behind and become losers. It is unlikely that any international treaties will be introduced within a reasonable time frame.
Generative AI – A threat?
Various impressive AI solutions have been developed in recent years. Generative AI poses several risks, including misuse, misrepresentation, bias, lack of accountability, intellectual property infringement and economic disruption. The emergence of AI systems could expose 300 million full-time jobs to automation, and two-thirds of US occupations could be exposed to some degree of automation. These concerns and doomsday predictions about AI have led to calls for caution and restraint. Some people, including some of the developers of these language models, believe that we are close to developing general artificial intelligence.
However, there are many reasons to question whether this is the case. Language models excel at language, which is a crucial step forward. This skill is necessary for translating between human and computer languages. It can bridge other algorithms and systems. However, rather than being impressed by ChatGPT, we may be concerned about the predictability of our language.
Are a few large companies controlling Generative AI?
Some voices have expressed concern that this technology is in the hands of a few large American companies running in large American data centres. At first glance, this may sound like a legitimate concern. However, it is important to realise that OpenAI was a relatively small, albeit well-funded, start-up. The cost of renting more than 10,000 GPUs to train the GPT-3.5 model for more than a month is $4.6 million. That’s a lot of money, but not more than a few companies around the world could afford.
Today, anyone can rent cloud-based resources. Training data is also freely available. Common Crawl is a monthly snapshot of the entire internet and is a free dataset that anyone can download. Of course, if you want to download a 400 TB dataset, you may need to buy some hard drives first! Although training a model is costly, it can be used on much simpler resources. Therefore, the competition to catch up with OpenAI and Google would be tougher than most people realise.
Open source models
Despite calls for a release freeze on new AI models, there are good reasons to expect very rapid development of these technologies over the coming year. In addition, there are already some promising models based on open source. Stanford University has developed a much smaller language model called Alpaca. It is so small that they distribute it as open source software that you can train and run on your computer. According to their comparison, their model is almost as good as GPT-4. Other alternatives are Koala, GPT4All and Vicuna. However, none of these licences allow the user to use the models commercially.
Recently, however, DataBricks released an open source model, Dolly 2.0, which users are allowed to use for any purpose and which they can train on any data they want. It can be downloaded here.
In a leaked document written by a Google employee, open-source models will ultimately win out over larger companies because they develop well-functioning smaller solutions with much smaller budgets and better performance3.
Generative AI – An opportunity?
It is important to realise that the economy usually adapts and grows in line with technological developments. This type of AI also offers many opportunities for growth and development. Goldman Sachs Research believes that generative AI could have a significant impact on the global economy. According to them, generative AI could lead to 7% growth in the global economy (i.e. GDP growth of almost $7 trillion) and also increase productivity growth by 1.5 percentage points over the next decade.
Advances in natural language processing enable generative AI to create text, images and other content that is no longer distinguishable from human creations. This will break down communication barriers between humans and machines. Goldman Sachs economists Joseph Briggs and Devesh Kodnani argue that most jobs and industries are currently only partially exposed to automation. They further believe that AI is more likely to complement than replace human labour.
Historically, jobs that have been replaced by automation have always been compensated for by the creation of new jobs. The majority of long-term job growth can be explained by the emergence of new professions due to technological breakthroughs. According to researcher and economist David Autor, 60 per cent of today’s workers are employed in jobs that did not exist in 1940. Technology-driven job growth has accounted for more than 85% of employment growth over the past 80 years. There is therefore no reason to take threats that AI will take over all jobs seriously. This is not how new technology has historically affected the world. On the contrary.
AI will not replace you. A person using AI will4.
Consequences
Advances in artificial intelligence will have far-reaching consequences for global IT development, healthcare and financial services. Generative AI has the potential to improve business workflows, automate routine tasks and create a new generation of commercial applications. It is already increasing the productivity of knowledge workers, accelerating medical research and increasing the efficiency of programmers.
Compared to the global software industry (a market worth approximately $685 billion), the total market for generative AI software is expected to reach $150 billion. Companies in various industries will also benefit when different types of generative AI technology are created and incorporated into current software packages and technical platforms. Applications range from increasing office efficiency and sales to design and manufacturing, improving patient diagnoses in healthcare, and detecting cyber fraud.
Summary
Rather than halting development, a responsible approach to AI development should focus on addressing the risks and creating sensible, balanced regulations. Organisations can ensure that design and training are carried out responsibly, establish guidelines and protocols, educate stakeholders, implement safeguards, and monitor and evaluate the performance of generative AI systems.
If we do this right while being mindful of potential risks, the positive effects of generative AI can yield positive outcomes for many and drive innovation and productivity improvements.
As history has shown, the creation of new jobs and industries often compensates for the job changes caused by automation. But change is always painful. Instead of fearing AI-induced unemployment, we should focus on the potential opportunities for new jobs and economic growth that AI can offer. This requires a balanced and sensible approach to the role of AI in the future economy and labour market.
Generative AI, like so many other technologies that came before it, certainly poses a threat with risks that must be managed. Critics are likely to exaggerate the risks, while optimists often have an unrealistic view of the technology’s possibilities.
At Gislen Software, we are building expertise in various types of AI. Please contact us for more information on how we can help you with AI solutions or any other type of software development!
Footnotes
1 Generative AI is a collective term for various AI technologies that are fundamentally based on deep learning – which can produce text, code and visual media by identifying patterns in text. They are trained on very large amounts of training data (often referred to as Large Language Models). GPT-3, for example, is trained on approximately 45 TB of text data.
2 The image of the lamp is by macrovector_official on Freepik.
3 The comment about the leaked document was added on 6 May.
4 The quote is used all over the internet. It is unclear who said it first.