How Artificial Intelligence (AI) and Deep Learning Impact the Future
An Interview of Dr. Maya Dillon, Head of Data Science, Global Centers of Expertise at Luxoft, a global technology consulting company focused on business transformation solutions for Fortune 500 companies.
By John W. Koon, Editor-in-Chief
Dr. Dillon provides support for data science solutions across Luxoft’s Lines of Business. She helps clients make sense of large, disparate data sources to extract real actionable insight. As a result, she helps businesses deliver new, differentiated products, driving competitive advantage. Dillon is a member of the Tech London Advocates organization and a supporter of The Royal Astronomical Society. She received her Doctorate in Astrophysics from the University of Warwick.
1. Can you provide an overview of what Artificial Intelligence (AI) and Deep Learning and how it will impact the future development of technologies? Please include an example or two.
There are two particular arenas of AI that are fascinating to me:
Healthcare: Diagnosis and Treatment of Diseases:
AI is now capable of diagnosing diseases with greater accuracy than human doctors by taking into account a larger number of factors. The use of AI in subsequent treatment is also compelling, particularly in the case of cancer. AI now supports everything from the identification of tumors, to implementing therapy, to aiding the excision of masses. Such methods are vastly improving the efficacy of current treatment plans, and consequently improve the quality and longevity of patients’ lives.
Automotive: Self-Driving Vehicles:
The key to a successful self-driving vehicle is ensuring the AI controlling the vehicle is constantly aware of the events occurring around it, in order to recommend or enforce a course of action.
However, it’s not just enough to know. Computer Vision AI has become ever-more accurate in labelling images in ‘real-time’, but the extrapolation of events into the future is still something that is being developed. For example, computer vision can enable the identification of a pedestrian next to the road vs a tree or signpost. However, having now identified the pedestrian, are they going to step/run out in front of the vehicle? Or do they look like they are simply continuing their trajectory down the pavement? That’s where computer vision needs to head next.
2. What is your view of future AI?
In the next 3-5 years, I think we are going to see an acceleration in terms of the accuracy of AI in all arenas. The bottleneck thus far has been the computational capability and processing of data, i.e. aggregation and mining. The math has been around for a long time. With the advent of new cloud computing technologies, i.e. cheaper storage, faster processors and a wealth of data in both structured and unstructured forms, there will be an explosion of realized potentials.
For me, what is important to note is we are still not at the optimal cognitive/intelligence computing capability – despite what you hear. True cognition (as we recognize it) is the capability of taking information, abstracting from it, and making a leap in judgment processes. Many AIs, as part of knowledge-based systems (KBS), can give you the impression that they have true cognition. They can answer questions, like, ‘Is this information important to me in this context? How does it relate to other topics? What options do I have and what are the likely outcomes?’
However, most KBS are highly contextualized. The domain-specific information they are trained on depends on the quality of human input. Think of chatbots for Robo Advisory or customer services support. There is no concept of intuition within these systems – the autonomous ‘desire’ to learn, and for the AI to ‘reflect’ on the ‘value’ of the information acquired – too truly problem solve! These are still very human concepts and processes. I think some of the leaders in the field of AI will make great progress in mimicking these concepts. For many end users in the next 2-3 years, it will start to feel as though we are actually talking to intelligent systems – but true cognition is a while away, yet.
3. Who are the leaders in AI today?
Everyone already knows them as household names: Google, Microsoft, IBM, Apple and Tesla.
4. Can you provide a brief overview of who Luxoft is? What are your main offerings and how you fit in AI today?
Our approach at Luxoft has been and always will be to help organizations solve important business problems. We look to domain experts to tell us what matters to them, and support by providing creative predictive analytics solutions through the help of our experts in AI, machine learning and data science. We build onto the software development heritage the organization has, bringing forth a brighter future.
We also work across multiple sectors, providing AI solutions in chatbot, computer vision, predictive maintenance, fraud analytics and blockchain.
5. What do you see as the main challenges of AI development today?
I think there are several issues:
- Lack of appropriately skilled people – there is a massive global staffing shortfall.
- Lack of organizational knowledge of what AI can and cannot do – we are still in the hype period as far as I am experiencing, and education is the key to deriving value from endeavors.
- Lack of knowledge in implementing tools and technologies – particularly in relation to legislation, regulations and compliance. However, this should not be seen as a barrier but as a rite of passage. The objective of these processes is to protect the rights of individuals and organizations. Therefore, pushing headlong into implementation without considering the overall positive and negative impacts is a part of the march towards progress – not something, that is impeding it.
Finally, regulatory processes always have, and always will, fall behind the pace of new technology development. The key is to ensure regulators understand what is happening in the market, so that new legislation can anticipate the development of new avenues for technologies in the nearer (and more “predictable”) short-term.
6. Many have raised questions that while AI can provide solutions in areas like healthcare and smart manufacturing, it can also help hackers to create smart hacking software. What needs to be done to safeguard this?
From my rather brief stint in cybersecurity and my continued involvement as an ‘active’ observer, the solution is threefold:
- Ensure that legislation and regulators keep up to date with technologies as much as is possible.
- Create processes to protect against security breaches within organizations due to negligence. Sadly, the biggest threat to privacy and security is human negligence perpetuated by poorly enforced processes, and not external threats – the latter which usually will gain a high profile through the aid of the media.
- You can’t anticipate and prevent every attack. What should be at the forefront of an organization’s mind is: it’s not a matter of if someone is going to break into my system, but what am I going to do when it happens.For example, for DDOS attacks, the objective is to recover control of your systems as quickly as possible. However, as most hacking is driven by the desire to acquire user information, the appropriate processes must be in place to ensure customers are made aware of the threat as quickly as possible (that their information may have been acquired), while supporting them to mitigate the subsequent fallout.
Of course there is a fourth, more innovative approach: the creation of AIs that tackle AI-driven malware itself. The AI arms race has begun! Are you ready?
About the company: Luxoft is a global IT service provider that delivers innovative technology solutions to multinational companies which enable them to meet evolving digital challenges. To this end, Luxoft combines strategic consulting with custom software development services which are supported by digital engineering centers. Luxoft has over 12,700 employees across 40 offices in 19 countries within five continents, with its operating headquarters office in Zug, Switzerland.