Artificial Intelligence (AI) has been mooted as the most transformative technology of the 21st Century as we allow computer algorithms to run more aspects of our lives.
Amazon Echo, Apple’s Siri and Google Home all demonstrate consumer appetite for AI-enabled products. Over the next few years, AI is expected to deliver driverless cars, better healthcare and truly smart homes.
The UK is rated in the top three countries for AI development and government figures suggest the UK’s AI industry could add £630 billion to the economy by 2035, increasing annual growth rates to more than during the boom decades of the fifties and sixties.
Professor Stephen Hawking has stated “once humans develop Artificial Intelligence, it will take off on its own and redevelop at an ever increasing rate.”
Below, we look at the achievements within AI to date and where it needs to develop to be truly transformative.
The Healthcare Sector has embraced AI as it looks to utilise the huge amount of data amassed; from pharmaceutical companies using algorithms to develop new drug hypotheses using data collected from previous trials to cardiograms predicting future heart issues from tracking devices like ‘fit bits’. An AI system developed at John Radcliffe Hospital has been shown to diagnose heart scans more accurately than heart consultants. Clinical trial results from six cardiology units are due out in 2018. If early results are confirmed, it is estimated the AI systems could save the NHS over £300m per year in cutting down the number of unnecessary operations and the treatment of people who had heart attacks following all-clear scans. Another AI development will be local surgeries increasing efficiencies through automated virtual doctor/patient consultations.
In 1997 IBM’s Deep Blue beat Kasparov at chess by the computer being loaded with the rules of chess and then calculating the best move by assessing each possible outcome. This is an example of cognitive learning. It has only been in the last few years that the AI industry has been able to add perceptive and manipulative learning developed through advances in machine learning techniques.
How does AI work?
Overall intelligence is the combination of tacit knowledge (intuitive learning based on perception) and explicit knowledge (based on learning).
Machine Learning is a machine’s ability to deliver on a task according to how it has been programmed but to also learn from that experience to complete additional tasks. Machine learning enables the machine to keep improving its performance without humans having to explain exactly how to accomplish all the tasks it’s given and is possible through the combination of algorithms and code within big data. Initially, items of data are labelled so computers can recognise them (e.g. an image) based upon the explicit knowledge provided. Machine learning is where the computer can take all the examples of labelled data and then identify unlabelled images through inference (tacit knowledge).
The neuro network technology enabling machines to use tacit knowledge has been available for 20 years but needed massive amounts of labelled data in order to be put into practice. The labelled data was produced through the mass adoption of mobile phones and the internet. For example producing huge amounts of images shared through social media or the internet providing large amounts of user data. This is extended to records now being kept on every day activities and events allowing machine learning techniques to capture intelligence from this mass of information on humanity. So people have not only constant information at their fingertips leading to a much wider knowledge base than before but also machine learning can make available knowledge before we have made the inference.
Will AI Take-Over and Lead The March of the Machines?
It is important to remember that although AI can start to make decisions and set tasks for itself, its general purpose and reason for existence is ultimately defined at inception. The next generation AI-powered applications being developed by IBM, Google and Apple use specialised AI code to replace the human element for specific tasks. These tend to be specific, almost tunnel-vision-like programs to perform dangerous, monotonous or time-consuming jobs.
Further Development Needed
An obvious concern is that all the above knowledge is based upon in the initial data entered. A little like Wikipedia; launched as a democratic encyclopedia based upon a large open source of information to produce more reliable findings. Wikipedia’s strength is also its weakness as it is open to human error. At the same time, if it became an aggregator of information and published the most commonly held view then the knowledge would be dumbed down and the advance of new information or thinking would be inhibited. The end result is that Wikipedia is not held as a reliable source of information. There is a lesson for AI as in a data driven society it is essential that the information is trusted and not subject to the existing biases in our society.
Another major issue for AI is an inability to generalise. Computer programs, even artificially intelligent ones, work far better as specialists. Much of human intelligence relies upon the ability to abstract from the specific to highly general information. This is essential to create a transferable intelligence in order to cover multiple areas of expertise or conduct multiple tasks. One proposed solution is that building a network of AI machines will develop a broad intelligence through a network of sub-programs. Artificial vision through image recognition, language through voice recognition, movement through robotics etc. A transferable intelligence relies upon being able to develop a broad, general knowledge AI, which is not possible today and depends upon Prof Hawking’s assertion that AI will self-develop with mass application.
Much work will be needed to achieve such a goal; neural networks typically have a few million artificial neurons compared to over 100 billion real neurons in a human brain.
One school of thought is that existing machine learning techniques, currently used to extract tacit knowledge from the mass of data, can be further developed to develop further inference from the mass of tacit knowledge generated. Drawing inference from inference.
Others believe this will require a reinvention of the recognition techniques in order to develop reasoning. The current direction of the AI industry is to try and leverage upon all the work that has been done and the inference techniques that have been developed rather than reinventing the wheel.
For AI to think like humans, it will need to add artificial emotional and social intelligence. This will be essential in order for AI systems to work alongside humans in a team on a collaborative basis. This will enable all parties to recognise what others are doing and also be aware of how certain actions could affect others in the team. The AI models will need to interact with different members of a team in different ways based upon that person’s understanding and role within the team. For example a doctor’s explanation of a particular diagnosis would be at a different level and use different language when talking to a colleague with a broad existing medical knowledge rather than to a patient.
What can we expect in 2018?
During 2018, we are likely to see more and more use of Artificial Intelligence in 2018 as it is introduced to augment human intelligence. This is where we start to use machines to complete the repetitive parts of our every day jobs, freeing our time for the more creative roles or areas that require expertise. A good example again is medical practitioners whereby AI will assess a patient’s medical history, test results and scans and present a list of most likely diagnoses to doctors. Doctors can then interpret the various diagnoses, identify the possible treatment options and communicate all the findings to the patient.
Existing AI cannot replace doctors but a world where doctors are more effective and efficient with the help of AI is within our grasp. Increased efficiency has been a long term target for the NHS.
Introducing AI through augmentation could also be significant in winning over the public with a lack of trust being a major obstacle for new technologies. If people see the benefits such systems can bring with minimal disruption then they are more likely to advocate further developments and usage.
Investing in Artificial Intelligence
If AI is to be the most transformative technology of this century, then many investors will be looking to get in early and try to identify UK companies capable of building market share in a fast evolving space. As always with technology, ease of use will be a key consideration. If an AI system is easily augmented with everyday practices, customers are more likely to buy it. Other small company factors include the usual essentials of a strong management team, an edge over competition and a feasible business plan with clear value creation milestones.
In conclusion, the more widely AI is used, the faster it will develop leading to significant advances. Many predict we will see greater technological change in the next decade than we did over the last ten years. When you consider the huge adoption of technologies since 2007, we better strap ourselves in for the AI-driven ride ahead.
The information in this website is provided by CSS Partners LLP. This website has been approved for the purposes of section 21 of the Financial Services and Markets Act by Charles Street Securities Europe LLP (CSSE), which is authorised and regulated by the Financial Conduct Authority. CSS Partners is an appointed representative of CSSE.
Any views or opinions expressed in this blog are those of the author alone, except where specifically stated that they are the views of CSS Partners LLP.