Insights from CWTEC 2024: Engineering AI for Critical Systems

Simon Rockman, Chief-of-Staff at Telet Research and CW Board Member, shares his insights from CWTEC 2024.

Catering for the intellectually curious is the home territory of Cambridge Wireless. In the days when Old Street was fashionable as ‘Silicon Roundabout’, those in the know appreciated that what went on there was about using technology rather than inventing it. The real invention happened in Cambridge.

Cambridge has a long history of real innovation and invention, often through spinouts from the university but sometimes because it is home to the kinds of companies that the intellectually curious want to work for. It’s the role of Cambridge Wireless to help those companies become more than the sum of their parts by cross fertilising the knowledge within them. There are three foci for this. Cambridge Tech Week, which is September, CW International Conference which will be held early next year and CWTEC which was last week. While the first two events span the rich and wide variety of technologies in Cambridge, CW TEC dives deep into one particular aspect.

The speakers looked at the mechanisms to build a better, safer world through highlighting the potential of AI to enhance diagnostics, resilience and security. Among the special guest speakers were: Chris Murphy, Regional CTO, EMEA, Viavi Solutions; Stephen Clemmet, CEO, Silogic Technology; Dr Camille Terfve, Partner, Mewburn Ellis; and, Cheryl Allebrand, Senior Consultant, Artificial Intelligence (AI) and Automation, CGI. They gave practical examples of what had been done in engineering AI within vertical markets from robotics to wind turbines, from telecoms to transport infrastructure. With a good mix of where the underlying technology has come from, where it is now, where it's going, and the technical, practical, and ethical considerations.

It may seem pedestrian that this year we had an AI theme. The event has a history of being future looking and AI is very much mainstream, but that is what made this year’s event powerful on the theme of “Engineering AI for Critical Systems” it wasn’t about what AI is, but what we’ve learnt from deploying it.  The event highlighted the importance of trust. Not just in the code, the choices systems make, but in the entire ecosystem. That’s not just what the hardware and software do but the people and cultures around them. The machine has to learn from the real world without opening the door for malfeasance.

How this is done was presented through a number of case studies, from a keynote speech delivered by Professor Amanda Prorok, University of Cambridge, entitled: “Learning to Communicate, Cooperate, and Coordinate in Multi-Robot Systems.” In this talk Prof Prorok explained why, although we all knew of highly automated warehouses like Amazon and Ocardo, they were vey much the exception rather than the norm. She explained what was necessary to have software agents which worked in an efficient and co-operative manner.

When mainstream media wants to illustrate AI, they will usually do so with a picture of a robot. That’s usually shoddy shorthand: robots don’t necessarily have anything with AI and of course AI does a lot more but in the case of Professor Amanda Prorok’s keynote the two go and in hand. Prorok, Principal Investigator, Prorok Lab and Professor of Collective Intelligence & Robotics at the University of Cambridge, looked at a data driven approach to getting robots – software and hardware – to do each other’s bidding.

Key to this is the Pareto front, software which rises to the top in terms of the speed and quality of the decisions it makes. The University used this for cooperative perception, coordinated path planning, and close-proximity quadrotor flight.

A key architecture decision in getting this to work was moving from a centralised to a decentralised model. This created the problem of how the software agents communicated. Draft Analysis was deployed to teach the agents how to communicate. Draft Analysis is a process built on modelling how one agent's observation affects the action of another agent, and this interaction. This is used to synthesise the messages that agents send each other so that they can coordinate towards solving priority applications. As with many things in AI the bigger the problem gets – the more agents you have – the more data you have and the better the result.  This allows flying drones not just to avoid each other, but when flying in close proximity to avoid the aerodynamic problems of disturbing the air close to another drone.

Perhaps the most extraordinary demonstration at CWTEC was Stephen Clemmet’s explanation of a real-world hack of a Wind Turbine. Clemmet is the CEO of Silogic Technology which uses AI in its arsenal of tools to provide cyber security for wind turbine operators. He’s yet to encounter a client who didn’t have something to worry about. Like most power systems, wind turbines use a Supervisory Control and Data Acquisition (SCADA) bus. Silogic mines the data on the bus to find vulnerabilities and uses AI to look at the pattern of the data. This helps identify the potential for a virus to cause a denial-of-service attack. If a wind turbine is taken off line it’s losing money in real time. The miscreant will then blackmail the owner. If a virus gets in the turbine, which could very well be in an area which is hard to access, it cannot be reset remotely.

The tools and the user of them understands that hackers play a long game, they may have compromised the system many months before they mount the attack. In the example Clemmet presented the hack went on for six months. The AI software looks for gaps in data which may just last a few seconds. By knowing what the data should look like all the time, and spotting something irregular, the crime can be averted.

Dr Camille Tergave of Mewburn Ellis looked at a different form of protection: AI in Health, Pharma and Biotech. Therapeutic antibodies are a relatively new form of treatment and they combine the development of affordable sequencing with AI to help identify proteins that will bind to maladies. In particular, Tergave looked at first line treatments for cancer. Her team has developed this algorithm that can identify mutations with small fragments of data and vaccines that target the particular cancers.

Engineering and ethics are not usually seen as bedfellows, but AI has changed this and the event gave space to step back and appreciate that while much of what is touted as AI, but is just a database look-up, we are entering a brave new world where machines are making value judgements. The maths behind what they are doing might be traceable back hundreds of years but the mantra of “technology will never move as slowly as it does today” is truer about AI than any area of technology we’ve yet seen.

There is an easy way to tell the difference between speakers who are pitching for business and those seeking to educate an audience: the questions and answers sessions.

Those speaking at the attendees don’t take questions. They set out their stall and leave it at that. Cambridge Wireless, doesn’t do this, events are heavily about interaction. The questions are intelligent and the answers more so. Discussions continue into the breaks and business is done over coffee and biscuits.

It’s events like CW TEC which reflect and cement Cambridge as a place where real technology is researched, developed and sold. There wasn’t a whiff of silicon development in Shoreditch and now there isn’t even a roundabout.