The Latest

10 May 2022: Microsoft Research has introduced an advanced prototype of PeopleLens for young learners with visual impairment at the University of Bristol. 

The solution uses augmented reality eyeglasses tethered to a mobile device to identify people and track their direction and distance from the user. Using artificial intelligence (AI), the solution registers people in the system through facial recognition and alerts the wearer in real-time by identifying the person and their distance and direction through spatialised audio. To protect privacy, facial images in the system are not stored as photographs but as vector numbers to represent identities. The technology is not yet commercially available, but does provide hints at what the near future will

Why it’s Important

Education is something of a laggard in the application of AI, especially in Western economies. 

However, innovations such as PeopleLens provide a glimpse (pun intended) of what is possible. Using AI in education is expected to grow quickly, but where and how it will be applied is as much a matter of economics as it is technology. 

The cost of AI at scale can be a prominent issue in this case. AI computation may be inexpensive in cases where requests are relatively small, but costs can quickly add up for applications that require millions or even billions of transactions. In addition, releasing new AI algorithms is still relatively expensive, due to the high cost of investment in research and design, as well as expenditures for the development of prototypes, complementary equipment and software. Hyperscale Cloud computing helps reduce these initial expenditures, but training is still required. 

Therefore, the business cases for an AI initiative must be carefully weighed against the potential future scale versus the value to individuals. In short, does it scale economically?

In a recent IBRS interview with an Australian Microsoft Azure specialist who developed an AI model to detect improper Microsoft Teams usage among students - such as cyberbullying, aberrant behaviour and inappropriate content sharing platforms - the transactional cost was not feasible, even with the aggregate value of securing children from harassment online. Since the Teams environment hosted hundreds-of-thousands of users, each producing scores if not hundreds of messages daily, the total cost of running the solution was not a viable commercial option.

In the case of PeopleLens, on the other hand, the number of transactions per individual may be relatively high, but the number of transactions as an aggregate is relatively low. As such, it is potentially an example of where the value returned is acceptable when compared against the cost. 

Who’s impacted

  • CEO
  • Innovation managers
  • Education policy strategists
  • AI solution development teams
  • Product research teams

What’s Next?

Industries that are planning to leverage AI effectively and at scale should ask for examples of how different AI-powered solutions are being justified.  

For most organisations, AI will be leveraged as features from within SaaS solutions, such as SalesForce's Einstein and Microsoft's use of GPT3 inside the PowerPlatform. 

However, for those looking to create new applications that leverage Cloud and ML capabilities, transactional volume should be carefully considered early in the planning stage to accrue the most value from the investments in research, design, development and production in the long run.

Related IBRS Advisory

  1. All Together Now! Hybrid Work, Technology, Diversity & Inclusion
  2. Innovation: Taking action in 2018