Should we use AI in cybersecurity? Yes, but with caution and human help

Image: iStock/Blue Planet Studio

Artificial intelligence is accelerated becoming the redeeming grace erstwhile it comes to cybersecurity. In a caller post, Reliance connected AI successful effect to cyber attacks 2019, by country, connected Statista, Shanhong Liu said: "As of 2019, astir 83% of respondents based successful the United States believed their enactment would not beryllium capable to respond to cyberattacks without AI."

SEE: Security incidental effect policy (TechRepublic Premium)

That startling statistic captured the attraction of Martin Banks, who, successful his Robotics Tomorrow article, What Security Privileges Should We Give to AI?, asked the pursuing questions:

• Are determination limits to what we should let AI to control?
• What information privileges should we entrust to AI?

How overmuch cybersecurity is harmless to automate?

Banks said AI is an fantabulous instrumentality for:

• Authenticating and authorizing users
• Detecting threats and imaginable onslaught vectors
• Taking contiguous actions against cyber events
• Learning caller menace profiles and vectors done natural connection processing 
• Securing conditional entree points
• Identifying viruses, malware, ransomware and malicious code

The takeaway is that AI tin beryllium a potent cybersecurity tool. AI exertion has nary adjacent erstwhile it comes to real-time monitoring, menace detection and contiguous action. "AI information solutions tin respond faster and with much accuracy than immoderate human," Banks said. "AI exertion besides frees up information professionals to absorption connected mission-critical operations."

Here's the tricky part

For AI to beryllium effective, the exertion needs entree to data, including delicate interior documents and lawsuit information. Banks said helium understands that AI exertion is worthless without this access. 

That said, Banks expressed a concern. AI exertion has limitations that stem from immoderate 1 of the following: a deficiency of strategy resources, insufficient computing power, poorly defined algorithms, poorly implemented algorithms oregon anemic rules and definitions. "Human-designed artificial quality besides displays assorted biases, often mimicking their creators, erstwhile turned escaped connected datasets," helium said. 

SEE: How to negociate passwords: Best practices and information tips (free PDF) (TechRepublic)

This speaks to Banks' interest astir the information privileges that AI exertion is entrusted with. AI exertion whitethorn attack perfection, but it volition ne'er wholly scope it. Anomalies are ever-present, and let's not hide AI-smart cyber criminals and their quality to subvert weaknesses successful AI systems.

There are much arguments for giving AI privileges successful cybersecurity than not. The trick, according to Banks, is striking a equilibrium betwixt AI (precise) and quality (nuanced) input.

AI with quality enactment is the champion solution

Banks said captious decisions, particularly those regarding users, should beryllium entrusted to a quality expert who has the last accidental successful however to proceed oregon what to change. "What if a idiosyncratic is morganatic and was flagged arsenic nefarious done a misunderstanding?" Banks asked. "That idiosyncratic could miss an full day's enactment oregon much depending connected however agelong it takes to place what happened."

The authors of the Immuniweb.com blog, Top 7 Most Common Errors When Implementing AI and Machine Learning Systems successful 2021, agreed. "AI has definite limits,'' the authors said. "In general, AI systems tin beryllium utilized arsenic an further instrumentality oregon astute assistant, but not arsenic a replacement for experienced cybersecurity specialists who, among different things, besides recognize the underlying concern context."

Banks, to marque his lawsuit for quality involution and power of AI processes, utilized a physical-security example: automatic information gates to restrict unauthorized traffic. "Grilles and gates support unwanted parties retired and let authorized unit entree to a property," Banks said. "Yet, astir high-security locations see quality guards arsenic an other precaution and deterrent."

"Gate systems tin analyse worker and vendor ID badges and marque a split-second determination astir providing entree oregon not," helium said. "But it's each data- and algorithm-driven. The quality guards stationed adjacent tin assistance guarantee the strategy isn't being exploited oregon making incorrect decisions based connected faulty logic."

Banks' statement is not whether AI exertion should beryllium deployed oregon not. His interest is astir the instauration connected which the exertion rests. If the instauration is built correctly with safeguards, we each volition benefit. 

Also see

• The information and privateness down IBM's Digital Health Pass (TechRepublic)
• Colonial Pipeline onslaught reminds america of our captious infrastructure's vulnerabilities (TechRepublic)
• How to go a cybersecurity pro: A cheat sheet (TechRepublic)
• Security threats connected the horizon: What IT pro's request to cognize (free PDF) (TechRepublic)
• Checklist: Securing integer information (TechRepublic Premium)
• Online information 101: Tips for protecting your privateness from hackers and spies (ZDNet)
• Cybersecurity and cyberwar: More must-read coverage (TechRepublic connected Flipboard)