Demystifying Artificial intelligence

Artificial intelligence (AI) is driving the fourth Industrial Revolution, building on the impact of steam power, electricity and digital technology — its impact will be widely felt across the whole of society. As this stage it is only in its infancy, this is the right time to start a national conversation around the potential, limitations and risks of AI in order to harness the power and have a real positive impact on society.

Every time you speak to Siri, use predictive text or scroll through recommendations on your Netflix or Facebook newsfeed, you are interacting with AI. AI is widely used in the world today but many people are still unclear about what it is and isn’t. So let’s discuss the terms often associated with AI to help to demystify it.

One definition of AI is the ability of a machine to perform tasks that require human intelligence. While AI research has been ongoing for decades, it has been the more recent advances in deep learning, processing power, accessibility of data and cloud services that have seen major breakthroughs in AI to the point where these systems are now highly effective and we see rapid deployment globally.

Science fiction is littered with tales of intelligent machines from 2001: A Space Odyssey to Star Trek and many more. In these storylines, the machine has intelligence equal to or surpassing humans; consciousness and the ability to learn, solve problems and plan for the future. This form of AI is known in the scientific community as “generalised” (or strong) AI. Some scientists believe it could be decades before generalised AI becomes a reality, while others doubt it ever can. Super AI, where AI surpasses complex human intelligence, is still purely speculative.

AI systems we interact with today are defined as “narrow” (or weak) AI that’s capable of performing a single or limited number of tasks. Within their field, these narrow AI systems are powerful, can replicate human performance, and in many cases even outperform humans. But it is worth noting that once these systems are presented with a situation that falls outside their learned space, they fail.

AI is powerful, prevalent and will continue to transform how we live and work into the future. That’s why ethical approaches to AI are needed and are in the process of being regulated, via the EU AI Act, for the benefit of society and to build trust in AI systems. These ethical approaches include data privacy protections and governance and transparency so that the outcome and decisions of AI can be easily explained, as well as making users aware of when they are interacting with an AI. Ethical AI also addresses the important issue of bias mitigation.

There has been a lot of discussion recently about bias in AI. This is when the system delivers different or poorer results for certain demographics based on gender, culture, age or race. In June, Microsoft also pulled from the market their sentiment and facial recognition systems due to the risk of bias. Stanford researchers recently discovered that voice AI developed by Microsoft, Apple, Google, IBM and Amazon returned significantly more errors for Black speakers than white speakers, likely due to an over-representation of white speakers in the data used to train the AI models.

Biased systems

The consequences of bias on the individual are significant, wrong and dangerous. Can you imagine a child using AI in the classroom getting told they’re wrong when they’re right? Confidence, skills acquisition and future opportunities are all negatively impacted by such biased systems. Mitigating such bias can be achieved through deliberate efforts in the construction of balanced datasets, representing the population who will use the technology; care must be taken in the labelling of the data and if using historic data to ensure any legacy human bias is removed, as well as in the training and evaluation of AI models. Decision making, free from human or systematic bias, is a prospect that will have huge societal benefit if prioritised in all AI deployments.

The positive impact of AI is already widespread. AI in healthcare includes the ability to perform medical imaging analysis to aid in the detection of common cancers and abnormalities; accelerating drug discoveries; robot-assisted surgery; predicting the likelihood of diseases, to name a few. AI is also being used widely to address the climate crisis by helping to improve the current understanding of how the climate and world is changing as well as contribute to solutions to the crisis. In education, AI is being leveraged to aid teachers in the classroom to help combat high student-teacher ratios and to deliver more personalised learning and assessment for better student outcomes.

Other hugely impactful effects AI has on business is in helping to build product innovations and efficiencies to help businesses compete globally. The impact of AI can already be found across almost every industry today.

There is a clear opportunity for Ireland to become a leader in advocating for and adopting an ethical approach to AI that puts humans first and is good for business. Building such a trusted brand for Ireland in a fast-paced AI world will help differentiate us in the market, as well as helping to ensure we mitigate risks to maximise the benefit of AI in our daily lives as individuals and communities across the country.

Patricia Scanlon is Ireland’s first AI Ambassador