A featured contribution from Leadership Perspectives: a curated forum reserved for leaders nominated by our subscribers and vetted by the Construction Tech Review Advisory Board.
Telefónica
Good+Greenai4business+4good Towards A Sustainable Ai Future
The use of artificial intelligence is increasingly penetrating our businesses and lives to such an extent that it is not enough anymore to look only at the business opportunities, but it becomes imperative to look at it from a 360 perspective. Apart from business opportunities, we also need to look at social opportunities, and make sure that the AI we use respects ethics and the environment.
AI4business
Many reports highlight the enormous business opportunity for artificial intelligence. PWC estimated in 2017 that by 2030, AI could generate almost $16 trillion annually. Popular business applications include chatbots, process optimisation, fraud analysis, diagnosis, facial recognition, content recommendation, to name just a few. Almost all large enterprises have set up AI or big data teams to create value for their businesses, either by optimisation or by creating new products and revenues.
AI4good
But there is not only commercial value that can be created with artificial intelligence. There are many applications that can help contribute or monitor the sustainable development goals of the United Nations. Big data and AI can be used to alleviate the impact of natural disasters. It can help predict and control the spread of pandemics, which has been important for the COVID-19 crisis. It can also help map out forced migration flows such that humanitarian organisations and governments can better ensure that migrants have access to health and education systems and it can help to fight climate change by monitoring and estimating CO2 and equivalent emissions and air quality in large cities.
goodAI
So far, we have seen the positive side of AI: social and business opportunities. But anybody who is attentive to the news will also be aware that there are potential negative consequences of this powerful technology.
Who hasn’t heard of AI systems that discriminate against coloured people? That they are black boxes that nobody can understand. That they increase social inequality by excluding minorities from essential services. It is not that organizations design AI systems to manifest such undesired behaviour, but oftentimes it is a side effect of how current AI systems work. They are based on machine learning, and in particular, on deep learning that takes as input a huge amount of data that the algorithm extracts patterns from. Patterns that allow the system to recognize cats, faces, cancers in X-rays, and even to automatically translate between hundreds of languages. Such systems learn from data, and if this training data is biased or not representative for the target audience, unlawful or undesired discrimination might happen. Moreover, deep learning algorithms are complex structures with hundreds of layers consisting of millions of nodes and weighted connections that no person can understand well. A few years ago initiatives have started to avoid such undesired negative consequences through the adoption and implementation of AI Ethics principles, including the European Commission, who is now preparing the first horizonal regulation of the use of AI.
By Considering All Those Aspects Of The Use Of Ai (Business, Social, Ethics, And Green), It Is Possible To Maximise The Value Of Ai While Reducing Its Unintended, Negative Consequences, And Build A Sustainable Ai Future
greenAI
The other important aspect to consider during the use of AI is its carbon footprint. Indeed, AI can help the fight against climate change, but it is also part of the climate problem. Large natural language processing models such as GPT-3 that process 1.5 billion parameters have a significant carbon footprint. Studies showed that training a single model once, had the same carbon footprint as five cars during their full lifetime. The training process of the GPT-3 AI model came with an electricity bill of $12 million. While this currently only happens with very large models, one has to consider that this is the cost of one training session of one model. Before an AI model works well, many training sessions are needed, and it is expected that in the future large AI models will operate in almost any large organization. To deal with this unintended, negative consequence of AI, research activities have started to systematically measure the energy consumption of AI algorithms and aim to come up with tools for estimating energy consumption and guidelines for programming and selecting AI algorithms that take--by design--energy consumption into account.
By considering all those aspects of the use of AI (business, social, ethics, and green), it is possible to maximise the value of AI while reducing its unintended, negative consequences, and build a sustainable AI future.
