Artificial Intelligence: Fascinating Opportunities & Emerging Challenges
Why I Selected this Topic
I selected this topic (https://soundcloud.com/bridging-the-gaps/artificial-intelligence-opportunities-and-challenges-with-professor-bart-selman) because of the role AI is playing in recent years in both the defensive and offensive aspect of information security. While defenders, e.g., manufacturers of antivirus are deploying AI to better detect malicious software using non-traditional methods, attackers are also using the same technology to deploy attack payloads in ways capable of defeating defense systems. I want hope to have deeper understanding of the workings of AI through the podcast.
About the Researcher
Bart Selman is a professor of computer science at Cornel University, fellow of American Association of Artificial Intelligence and American Association for the Advancement of Science and the president elect of the Association for the Advancement of Artificial Intelligence. He has special focus on computational and representational issues. The areas he has worked on include tractable inference, knowledge representation, stochastic search methods, theory approximation, knowledge compilation, planning, default reasoning, and the connections between computer science and statistical physics. His current research areas are fast reasoning methods, multi-agent systems, integration of learning and reasoning techniques and analysis of large linked networks
What the discussion is about:
The discussion is about researches into artificial intelligence and the fascinating capabilities that have been developed/achieved in the field. It also focused on the challenges of AI, one of which is the problem of explainability and lack of overarching regulation and global standardization of AI programmes. The problem of explainability seeks to bring transparency into the outcomes/results of AI processes as there are currently no way for machines/AI programmes to provide reasoning behind their results, which could create an issue of trust in the results.
I learnt the following from the discussion:
Research in AI has been on for about 60 years, however real progress began at the end of the 1990’s
A revolutionary landmark was achieved in the AI field in 2011/2012 with the deep learning computer programme. Deep Learning (DL) enabled computers to acquire different capabilities in specific areas through training on hundreds/thousands of samples, from which they are able to discern other samples (or perform other activities) they had not been trained on. Deep leaning is the mechanism behind autonomous cars, speech/image recognition, etc. Examples of the application of DL include language translators, deep blue machine the defeated Kasparov in the game of chess and the question answering IBM Watson computer.
The problem of explainability, the inability of AI systems to generate reasoning behind their outputs, which poses the issue of trust and ethics on the outcomes from AI programmes. What I make of this is, for example, is the if AI is used to digital evidence analysis, it will be difficult for this to be admissible in the court of law. However, research is ongoing to enable AI systems to generate reasoning behind their decisions.
Artificial general intelligence, which is focused on enabling AI system to act with flexibility similar to those of humans and adapt to different situation rather than focusing on specific areas such as self-driving cars, language interpretation, etc. Only little strides have been made in this area and it is project that it might take up to between 30 to 100 years for breakthroughs in this area of research.
There is a need to regulate AI or certify AI capabilities to meet certain standards to ensure ethical and trust standards are at acceptable levels. Regulation and cooperation between governments and researchers will also ensure that AI is not employed for negative purposes such as autonomous weapons
In conclusion, deriving from my reason for selecting this topic, I have found out that AI is a dual-use area, while, using deep learning algorithms, cyber defenders can make machines learn and identify malicious traits, the “bad actors” can also use the same techniques to learn how to circumvent deployed security mechanisms.