Dr Guru Banavar, VP of Cognitive Computing at IBM Research, gave the 2017 Turing Lecture - an overview of cognitive computing - to packed audiences in London, Cardiff, Dublin and Belfast. BCS editor Justin Richards MBCS summarises some of the key points from this prestigious lecture.

Dr Banavar began his talk on cognitive computing in a showman-like manner by screening a trailer for the new sci-fi–horror film, ‘Morgan’, which had itself been created by IBM’s own AI system, Watson, after first ‘sending Watson to film school’. It had analysed the whole film and had suggested clips to be included and their sequence.

He went on to say: ‘Over the past few years we’ve witnessed the establishment of a new era in computing - the age of machine learning. And, as we move into this new age, the resulting technical, professional and societal changes will be profound.’

Rounding off his initial summary, Dr Banavar said: ‘This means having a very different relationship with machines. We need to start getting used to having machines with us, to having natural interactions with them, and get used to the idea that they’ll be doing a lot of tasks in every aspect of our lives.’

Banavar believes that, currently, the most exciting aspect within the field of AI is looking at how machines can work alongside humans.

Banavar thinks that the next revolution will involve the augmentation of the human condition through machines. To Banavar’s mind this will be of such magnitude that it’s on a par with the previous industrial revolutions. Bearing in mind that each revolution usually lasts for 50 -70 years, Banavar believes we’re at the beginning of a new era - the ‘cognitive era’.

We, as humans, are now getting overwhelmed with respect to our cognitive capabilities, trying to understand all of the data and knowledge around us, and making the right decisions about our jobs and daily lives is getting harder so we need to augment our cognitive skills with the cognition of machines.

Mountains of data

It’s predicted that by 2025 there will have been a total of 165 zettabytes (one zettabyte is approximately equal to one billion terabytes) of data generated. Of all the data generated so far, 90 per cent has been generated over the last two years. Most of this data is unstructured. All this raw data is just the beginning; we need to curate this data to make the most of it, but extracting knowledge from it is proving difficult.

Big data’s true value is in the embedded, multi-media context, but we have to first link the data, which is a massive job, and then perform a full contextual analysis. However, the process is complicated. This analysis could be done with the help of computers.

For example, holistic data from the entirety of environmental and biological sciences could have great impact on health. But we have a long way to go in understanding this and making useful decisions without the full knowledge of what we have.

Banavar sees such machines as tools to help, to augment human intelligence, through cognitive computing, and sees systems like Watson as being more akin to research assistants. The work IBM is doing now is in developing a real partnership between humans and machines; developing a reasoning machine. He believes that once it’s fully realised, cognitive computing will be to decision making as search engines were to information retrieval, which we all depend on today. He thinks we will depend on these cognitive machines in future.

With Watson they had to build large scale systems, using high performance computing (HPC) techniques to speed things up. They had very specialised hard and software systems, just as they did with the original ‘Jeopardy!’ programme back in 2011. Fast forward, and Watson can now easily learn Japanese, but not sarcasm or humour.

Data is becoming essential to value creation in every industry, whether it’s in water quality, food safety, pollution mitigation, seismic exploration and in drug synthesis. IBM has been able to augment experts in these fields with its AI systems.

Cognitive applications are assisting in every aspect of life - health (cancer and diabetes), education, transport, sports, and even tax preparation (H&R Block). In the case of diabetes research, they created an algorithm to predict sugar levels falling below a critical level. IBM have been able to build a very sophisticated system for healthcare because the data is so rich.

In fact, Watson recently helped to identify the top ten genome types involved, to help combat Amyotrophic Lateral Sclerosis (commonly known as Lou Gehrig’s Disease, a progressive neurodegenerative disease that causes muscle weakness, paralysis and, ultimately, respiratory failure).

A vision of the future

In future, everyone who needs expertise will have a cognitive assistant. For example, in healthcare AI can offer protocol options for practitioners; in finance AI could enhance portfolio analysis and risk management, and in education it could deliver personalised programmes for students and teachers.

‘All these applications are built on top of this platform - Watson is no longer an individual machine sitting somewhere and answering questions, but is rather a set of components that are available in a modern cloud-based architecture, with APIs, where people can build their own applications, using models and data available in the cloud’ said Dr Banavar. ‘Or you can build your own data and models, and this is becoming the way in which the world is going to change applications in pretty much every domain. That’s where were headed’, he concluded.

Dr Banavar believes there’s a need, not just for a deep learning system, but for a reasoning system. However, brains work in the opposite way to machines. Machines ‘experience’ an object first and then take in the background, whereas the human brain takes in the setting first before focussing on the object. This further complicates the issue.

Wide gauge vs narrow gauge

There seem to be two schools of thought. Firstly, there’s artificial general intelligence (AGI), where machines can pretty much do everything that humans can do. Such machines would have to learn how to do more abstract things including using pictures, writing essays and developing negotiating skills to achieve professional competence. These machines would literally need to go through the entire educational system that we, as sentient beings, do.

Banavar thinks that instead we should define tasks that are relevant to the problems we want to solve, so we can then deploy machines of ‘narrow AI’ today, rather than wait years for machines to catch up with us. This narrower, engineering approach is more effective and is yielding real results.

Social and ethical questions

Banavar believes that there are three key principles, regarding the cognitive era, that always need to be kept in mind, if the technological community are to keep the public and business on their side. These principles can be summarised thus:

  • Purpose - to augment human intelligence, rather than replace it’;
  • Transparency - we need to be transparent about what we’re doing, namely to what purposes AI is being applied;
  • Economic opportunities - to enable skills and knowledge to perform the work that will emerge in a cognitive economy.

We need to ask how AI can be misused too, and look at the wider implications. For example, many people will have to reskill themselves to be able to work alongside machines; it’s probable that many more menial jobs will go.

Occupations will have to restructure to meet elastic demand; for example, banking changed considerably once ATMs came in to being, with remaining employees being more focused on building relationships with customers and in investment management. Apparently, there are now more employees in the banking industry post the arrival of ATMs.

Where are we on our journey?

The AI achievement line is advancing rapidly and exponentially towards the Turing model. For example, the AI community are currently aiming to be able to run a robot football team by 2030. However, the first robot comic is still a very long way off!

Banavar finished off by saying that: ‘we must develop AI systems in a responsible and enduring way. We need to understand that this is not a fad, but will be here for a long time. AI will benefit and advance all humanity, but it needs all of us to work together to realise the promise of AI going forwards.’

About the speaker

Dr Guru Banavar is vice president and chief science officer for cognitive computing at IBM and is responsible for advancing the next generation of cognitive technologies and solutions with IBM’s global scientific ecosystem, including academia, government agencies and other partners. He has also served on task forces to improve resilience to natural disasters. He holds more than 25 patents and has published extensively.

Of note

IBM have created a new organisation called ‘Partnership on AI’ where Amazon, Facebook, Microsoft, IBM and DeepMind/Google have partnered with six independent non-corporation organisations, including ACLU and Open AI, to advance the public’s understanding of AI, formulate best practices for ethical AI, and to serve as an open platform for engagement on AI’s influence on people and society.