State of AI: Beneficial, but challenges persist
- By Matt Leonard
- Mar 30, 2018
What: “Technology Assessment: Artificial Intelligence: Emerging Opportunities, Challenges, and Implications,” a report from the Government Accountability Office
Why: Because AI presents relatively unknown opportunities, challenges and risks, the comptroller general held a July 2017 forum with AI experts in multiple fields to get a better understanding on where the technology is headed.
Findings: The report summarizes ideas and themes that emerged in discussions on AI in cybersecurity, automated vehicles, criminal justice and financial services. While there are many benefits to leveraging AI in these fields, like the potential to make cars safer or reducing human workload in IT offices, concerns over cybersecurity, privacy and civil rights must also be addressed. Other challenges include the amount of reliable data needed to train an algorithm, the computing and human resources required for AI applications, and the legal, regulatory and ethical issues that should be considered for using it.
Although many potential benefits were discussed, some forum participants noted that impacts may be difficult to measure and unpredictable. It may be more suited to fields like cybersecurity where the systems themselves generate so much data that AI can help determine normal and abnormal conditions, without less chance of introducing bias.
To advance AI, the report states, policies related to data sharing, security, regulation and risk need further attention, and research should focus on the following:
- Establishing regulatory sandboxes where AI products can be tested
- Developing high-quality, labeled data that can be used with AI to produce more accurate outcomes
- Understanding the implications of AI on jobs of the future
- Exploring computational ethics and explainable AI, in which systems can reason without being given explicit instructions and make adjustments for the future.
Verbatim: "The baseline is current practice, not perfection, and … the goal should be to become less biased and more accurate, not perfect."
Read the full report here.
Matt Leonard is a former reporter for GCN.