IEEE Software Magazine Nov/Dec 2022

 

Main points to the following articles in IEEE Software Magazine.  November/December issue was highly focused on comparing AI with Software Engineering.


An AI Engineer Versus a Software Engineer

"Looking at the generalized skill summaries of AI engineers and software engineers, we can observe that software engineers are expected to understand the end-to-end software development and sustainment lifecycle, and AI engineers are expected to contribute to that lifecycle by developing highly specialized components which take advantage of AI."

"The software engineering community needs to be realistic in judging the demand, refrain from turning all software engineering programs into AI engineering ones following the hype, and continue to refine their understanding of the overlapping and unique skill sets of both software and AI engineering disciplines."

Unlock the Business Value of Gamification

"Although gamification has been proven to have positive effects for software business and software developers, other methods for gaming in SE have also been developed. The most remarkable of these are game based learning (GBL) and serious games."

"Serious games are being widely used as a powerful tool to train users in virtual environments where making mistakes poses no risks, such as surgery practice simulation tools for medical education or flight simulators for pilot training."

Gamification Define the game objectives Delineate target behavior and metrics Describe the players Devise activity cycles Do not forget the fun Deploy the appropriate tools

Positives Stimulates connection between theory & practice Encourages participation Encourages learning from mistakes Gain experience without on-the-job risks Scalable

Risks Declining productivity & morale if cheating allowed Declining productivity when novelty wears off User Privacy User Personalities Demotivating Reward system Too much focus on speed rather than quality

Realizing the Potential of AI

"...when you initiate a project, it is nigh impossible to say whether a satisfactory solution can be found within given resource limits."

"To make matters worse, the skill sets needed in this discovery or the R&D phase are not the same as for software development, but instead, R&D is usually done by data scientists who, when they have created a working prototype, need to hand it over to the engineers who will build the production system.  Oftentimes, there are different coding practices, different frameworks allowed...frameworks are not 100% deterministic, new bugs are introduced in the redevelopment, and so on."

"If looking at state-of-the-art research on topics related to AI, the primary focus is on obtaining efficient computation or understanding the AI modeling principles (such as explainable AI), and there are many promising prototypes using AI technologies."

"Unfortunately, our research shows that the transition from the prototyping and experimentation stages to the production-quality deployment of ML/DL models proves to be a significant challenge for many companies. Though not recognized by everyone, the engineering challenges surrounding ML/DL model deployment are exceptional."

"And even if companies are equipped with skilled AI experts, these alone are not sufficient for building highly complex, software intensive, and AI-enabled systems that scale in domains that might be subject to safety-critical regulations.  Instead, there is a need for interdisciplinary teams that include AI  expertise as well as data science, domain knowledge, and in particular, software engineering (SE) expertise."

"As we work a lot with the embedded systems industry, where companies typically have hundreds, thousands, if not millions of devices in the field, we see a clear interest in using these devices in federated setups as the computational resources in these devices are not always fully engaged."