Every company now wants to be an AI company or embed AI into their offerings.
While the maximum potential of AI has not yet been tapped, there are many innovative as well as controversial applications of technology that have been tried and some also in production.
We are already seeing simple applications of AI to make things easier for the humans using these solutions – from predictive text while composing messages or even documents (this post has been completely hand crafted and not generated by a machine 😊), or chatbots that do not get confused when the response does not fit one of the options proposed, to more complex applications such as self learning robots in rescue and assistive situations.
Many teams have been experimenting with revised ways of working to structure the software and solutions development processes.
Some of the challenges and proposed approaches are highlighted in these posts by Anil Sane on the aithoughts.org site.
https://aithoughts.org/aiml-solutions-security-considerations/
https://aithoughts.org/ai-and-aiops-a-perspective-for-it-services-industry/
The emergence of data science as a specialization has also meant that the software development lifecycle needs to acknowledge this and incorporate it in the flow.
Governance aspects of AI based solution development are also undergoing changes.
Some inherent issues related to AI based solutions – such as ethics, has been explained in the post by L Ravi in https://aithoughts.org/ai-ethics-self-governance/, also need to be considered as part of the overall approaches and be reviewed
The unpredictability of real time usage situations adds to some of the complexity of the algorithms to be implemented. It is no longer just an algorithm that processes data. The data determines or influences the behavior of an algorithm or even choosing the appropriate algorithm to be chosen at runtime.
Additional ‘-ities’ are also emerging for AI based solutions such as ‘explainability’ as non-functional requirements that need to be considered.
There are some topics related to the overall solutions supply chain that are still being explored.
Some of these, such as the bias in the learning data, aim to improve the overall quality of the decisions derived or suggested by AI systems.
As with any emerging technology, there are some bad actors who keep looking out for holes in the solutions to be exploited, that affect the social good.
Deep Fakes are a good example of advanced algorithms being used to mislead or even trigger anxiety or unrest in individuals or communities.
With the increased interest in the metaverse and related business, information is going to get even more distributed and fragmented.
This would then mean that any solution designer needs to think of ecosystems and not just point solutions.
We have already seen the advantage of this approach – such as using location services or map services offered by some companies being embedded in solutions delivered by many businesses.
Thinking of the ecosystem, one must consider the full supply chain and the potential for benefit and fraud at every stage.
These include challenges related to the bias in the learning data, data poisoning, deep fakes and compromises to the edge devices or sensors.
A recent report by CB Insights identified the following as emerging areas of concern and action, when it comes to AI. Here are the highlights, with my own thoughts also included.
-
Synthetic data for privacy. This could also be very useful to increase the confidence in testing, where there is a constant struggle to share the real data between the developers and data scientists. It is not as simple to generate synthetic data and there are many considerations that go with it – to ensure adequacy, fairness [no bias] as well as constant validation of the neutrality of the data. We are used to capturing data mostly on the positive results. For example, We need to understand the patterns related to rejected parts during the manufacturing process, and that is another potential application of synthetic data – to generate images from quantitative or descriptive data that might have been captured in various analysis reports
-
Edge AI: embedding AI into the chips that would be the sensors [IoT devices?]. In addition to these being secure and immune to noise, some smartness would also be needed to ensure that these entry points are trusted
-
Policing or protection in the metaverse. While one may desire to have self regulation in the metaverse, one cannot wish away the need to have some mechanism of policing – essentially to create deterrents for abuse. Explainable AI and other such principles are useful in a reactive situation, but what is more important is to have some proactive mechanisms
-
Eliminating Deepfakes. We already spoke about this earlier in this article. When deepfakes are machine generated, the speed and volume could be a major challenge to combat
-
The report also talks about augmented coding. We are seeing tools from multiple vendors that are embedding intelligence in the development environment itself. For teams and organizations to learn faster, there would be a need to [selectively] share the learning across users. The question on how to tell the machine what may be shared and what may not be is another area that needs to mature
-
The next level of evolution of conversational AI, is to be able to support omni-channel interactions and multi modal AI, that can understand and process concepts from video, audio, text, images etc. this may be considered as the next evolution of natural interfaces and interactions, beyond just the spoken or written language
-
Black box AI – or end to end machine learning platforms would become the preferred option for enterprises to accelerate the adoption of company wide solutions
As seen from the above, the AI based solutions space offers enterprises unprecedented opportunities as well as unforeseen or unforeseeable complexities and challenges as the ecosystem also evolves.
In future articles, I intend to go deeper into some of the above trends and associated techniques or tools. The focus for me is to not lose human centricity when we embed more and more intelligence into the machines.
If you have any specific topics that you would like covered or questions to be answered, do reach out.