In a recent blog we briefly discussed the impact AI is having on our work lives – particularly in the non-technical careers. In this blog we will take a closer look at what AI is and some of the issues and challenges it presents.
Because AI is currently attracting the attention of many companies and receiving vast sums of investment dollars, new developments are occurring almost hourly – in the time it takes you to read this blog there will have been significant new developments. It is impossible in a blog like this to do justice to the subject, so we will focus here on just a few of the areas of concern being raised by these developments.
Issue number 1:
What are our moral responsibilities with this new technology?
As a society, we have been discussing artificial intelligence for a long time – at least since the early 18th Century with Leibnitz, and more recently, in the 1950s, with Alan Turing. But now we have the technology to actually implement these ideas.
At an operational level for business use, AI is a set of technologies that are based primarily on machine learning used for data analysis, predictions and forecasting, object categorisation, intelligent data retrieval, etc. Perhaps the most useful way of broadly categorising types of AI is by what the machine can do. Everything that we currently call AI is considered “narrow” intelligence. What we mean by this category is that it can only perform narrow sets of actions based on programming and training. For example, a system that is used for object classification cannot be used for language processing. Don’t be misled by this term though – the applications may be narrow, but some amazing actions are already being achieved.
The next category, artificial general intelligence (AGI) would be able to sense, think, and act just like a human. But importantly, AGI does not currently exist. In theory, the next level again, artificial superintelligence (ASI), is where the machine would be able to function in all ways superior to a human. Whether or not this is even possible (or desirable) is a question open to much debate.
It is critical that this debate be comprehensive and transparent to ensure the general public is accepting of these developments. Even existing narrow intelligence applications are having a major impact on society – even if most people are not fully aware of them.
Issue number 2:
AI systems learn and improve through exposure to vast amounts of data.
How this data is being acquired is generating much debate. Data mining, as it is called, collects, quite literally, every form of data available on the internet, and the people who generated and own this data are not necessarily even aware that their data has been “mined”. The issue of most concern is when the data mined is copyrighted data. Artists, musicians and other creators rely on copyright laws to protect their livelihood. Currently, they are not being compensated for the “use” of their data. How do we protect these people?
To make matters more complex, these same artists are more frequently utilising AI to generate their original works. How does copyright law apply to these new types of products? In some cases, a human contributes to the creation of the work, but in other cases, AI generates the whole product. For example, an AI generated portrait titled “Edmond de Belamy” was recently sold at a Christie’s auction in New York for $432,000. Who owns the copyright on this painting? There is an increasing amount of music being released that has been generated by AI – the same copyright question applies.
The generation of new laws is a long, slow process – it will never keep up with rapidly developing AI technology. Existing Australian and US laws require human authorship for copyright protection. This means that purely AI-generated works cannot be protected by existing copyright laws. But what about those cases where AI has only contributed to the end product?
There are many, many questions, but, so far, not too many answers.
Issue number 3:
The environmental concerns.
AI requires immense amounts of power to collect, store, and process data. These massive energy needs are driven by:
- The intensive computation required for training and running large language models and image generators.
- Data centre demand is driving exponential growth in data centres, which need power for servers and for cooling.
It is estimated that by 2030-2035, data centres could account for 20% of global electricity use. Even now, only a handful of organisations (Google, Microsoft, Amazon) can afford to train large-scale models due to the immense costs associated with hardware, electricity, cooling, and maintenance.
The most important ways that AI is impacting the environment are:
- The very high electricity usage – often fossil-fuel-based electricity.
- The very significant water usage for cooling systems.
- The short lifespans for the GPUs and other HPC components is generating significant electronic waste.
Reducing the environmental impact:
- Ironically, AI itself can be utilised to reduce this environmental impact.
- Advancements in hardware will play a part.
- Distributing AI computations across different time zones to maximise renewable energy availability.
So once again we see that the introduction of radically new technology generates both up-sides and down-sides. Our responsibility is to maximise the former and minimise the latter.