AI systems is on a fast pace, many AI based systems and tools are already emerging or in development stage and soon to hit the market. The business for AI looks exciting, however there is a dark side which is also emerging – the fear of ‘bias‘ that is overlooking these AI based systems.
Industry experts argue that an element of human bias would exist, and such bias will be ideally in the form of :
- Algorithms that make up the AI system
- Data on which the AI system is built
- Teams that build the AI systems
The perils of a bias AI program may lead to the entire focus shift to incorrect or wrong priorities or the system throwing up wrong patterns and the real threats may just fly out. In order to build a good AI system requires volumes of training data to learn, and deeper levels of trust so that the system is fair and unbiased.
The role of AI systems can be significant in the following ways:
- AI’s predictive analytics might render early warnings of cyber attack
- Early intrusion detection and monitoring
- Reinforcing strategic and tactical planning of cyber operations
So what are companies up to : Engineers at Microsoft are gearing to mitigate AI bias. What-if-tool at Google, AI Fairness 360 toolkit at IBM, Fairness flow at Facebook, Fairness tool at Accenture. Facebook is in collaboration with The Technical University of Munich (TUM) to support the creation of an independent AI ethics research center. Technical University of Munich is one of the top-ranked universities worldwide in the field of artificial intelligence.
Cyber attacks on AI
AI has enhanced security, but it comes with increased cyber threats. With AI getting adopted in almost every domain, the rise of AI cyber-attacks is also is on the rise. Common targets are AI botnets the cyber attacks being on data and vulnerable devices. Until last year, there has been a massive 20,000 plus botnet attacks on word press sites.
Vulnerabilities will still exist even with AI. AI systems can still be compromised and go undetected, as AI systems will be build for particular deductions and decisions are not always immediately clear to overseers. Most cyber attackers maintain a low profile and harder to detect, they get and manipulate data slowly without anyone noticing it. Technology giants like Google, Face book, Apple jointly have formed the Partnership on AI in 2016 to encourage research on ethics of AI including issues of bias