Formative artificial intelligence (AI) is an umbrella term given to a range of emerging AI and related technologies that can dynamically change in response to situational variances. Some may enable application developers and UX designers to create solutions using AI-enabled tools, while others facilitate the development of AI models that can evolve dynamically to adapt over time.
"The most advanced can generate novel models to solve specific problems," notes Svetlana Sicular, vice president analyst at Gartner.
In its Hype Cycle for Emerging Technologies 2020 report, Gartner put formative AI in its list of 30 must-watch technologies due to its disruptive nature and potential business benefits, and the analyst firm isn't alone in its views.
"According to economists, formative AU technologies aid in the drop in price of prediction," notes Monmayuri Ray, a GitLab solutions architect. "Prediction is at the heart of strategy making for uncertainty and hence, businesses should care about these technologies [because they can support] cost-effective prediction and decision making."
Enterprises looking to explore the boundaries of AI should analyse technologies that fall under the umbrella of formative AI, advises Sicular. According to Gartner these are AI-augmented design and development, ontologies and graphs, small data, composite AI, adaptive machine learning (ML), self-supervised learning, generative AI and generative adversarial networks.
"Enterprises should care about these innovations because they represent new capabilities compared to those actively exploited in AI today," she says.
"For example, AI-augmented design is a creativity tool and an efficiency generator. Generative AI created original artefacts or reconstructed content and data; it's a new way of applying ML techniques compared to the currently widespread methods that directly extract numerical or categorical insights from data.
"Ontologies and graphs allow companies to look at the immediate context, industry/domain specifics and data connections compared to the established data approaches centred on hard facts. Small data solutions are very useful for a long tail of unique, differentiating use cases for which enterprise don't have enough data."
Each of these technologies can be used for a particular purpose, and now's the time for enterprises to consider which might be a good fit for their business.
Formative AI use cases
Let's take a look at a few use case examples, starting with adaptive ML; also known as adaptive automatic learning. According to Jack Watts, EMEA leader, AI at NetApp, this technology could outpace traditional ML due to its ability to reach more accurate outcomes while investing less time and resources.
"At the moment it's being explored across a number of industries. For example, autonomous driving could leverage real-time scenarios to make systems behave much smarter over time with as little human intervention as possible. In healthcare, diagnosis systems could incorporate specialist knowledge and enhance the auto-diagnosis behaviour."
For a long time the compute power required to run adaptive ML was the preserve of multinationals and academic institutions. What's exciting about the current situation, says Rob Lamb, client principal at Dell Technologies, is that the power necessary has never been so easily available, and at such an affordable price.
Ontologies and graphs
Once the realm of the big tech firms like Facebook and Amazon, ontologies and graphs have been around for over a decade but there's a growing interest in its adoption across a range of knowledge-based domains like legal, says Alex Smith, global product management lead for iManage RAVN.
"The technology's being used in applications like document management and collaboration tools to intuitively harness the knowledge and best practice that resides in documents across an organisation. It's helping taking the traditional search to a whole other level to facilitate genuine knowledge management by surfacing recommendations based on previous history, much like Amazon, and contextual answers to business issues and problems, similar to Google.
"There's recognition that adoption of this technology can enable enterprises to connect internal business systems, derive knowledge, proactively surface commercially impacting information as well as achieve standardisation across the enterprise."
Generative AI is another interesting formative AI technology, because it can generate data sets to meet specific needs or conditions that aren't available in existing, original data notes Dr Nicolai Baldin, CEO and founder of Synthesized.
"Researchers have already used this type of AI to generate x-rays showing different medical conditions, which in turn have been used to train learning models. Because the data is synthetic, they can create far larger training sets that can be used without concern for patient privacy.
"In my opinion, the biggest development in the next 5-10 years will be that the accuracy of the technology will increase exponentially because the world will be producing more and more data, which in turn will enable us to tune AI models much more comprehensively than currently possible."
Also worth highlighting is AI-augmented development. This will enable currently overstretched development teams to deliver new software experiences faster, as problems are spotted, flagged and fixed earlier in the application lifecycle.
"This is a key principle of the 'shift-left' practices that many modern enterprises are adopting to accelerate their digital transformations," notes Alois Reitbauer, vice president and Chief Technology Strategist at Dynatrace.
Leveraging formative AI technology
So, what are the key considerations when looking to leverage these technologies? Â Ray recommends developing and operationalising a business-wide AI strategy and starting small with your first formative AI implementation. Lamb adds that data is key.
"When trying to distil every transaction, measurement or human interaction to an essence of 0s and 1s it's important that organisations understand the use cases they need to leverage data sources for, as bolting on new data repositories each time a need arises doesn't scale.
"A data strategy with a robust but flexible foundation is key, along with adherence to common standards and governance," he concludes.