Businesses must be more aware of the data requirements for artificial intelligence (AI), and use this period of focus on AI risks to prepare for the quantum computing ‘threat’.
That’s according to John Roese, global chief technology officer (CTO) at Dell, who shines a light on the main challenges businesses face when adopting AI models, and the lessons they can learn from the deployment of generative AI.
Roese acknowledges the computing bottleneck associated with training AI models, but denies this is the main hurdle holding businesses back when deploying the technology.
“The bigger issue for an enterprise use of a large language model (LLM) is in order to train it, you have to have access to the right data and provide the data to the training infrastructure,” he tells ITPro at Dell Technologies World 2023. “Most customers have not done enough work on their data management.”
Stripping out bias
As an example of good data management, Roese cites Dell’s work over the past four years to eliminate non-inclusive language from its content library and internal code environment. These include labels like ‘whitelist’ and ‘blacklist’.
If an LLM was trained on the firm’s content repository, Roese explains, it would be unlikely to not incorporate the biases of these words. Firms that don’t curate data before using it to train AI models intended for products such as chatbots could inadvertently create services that reflect an inherent racism or misogyny, as demonstrated by Microsoft’s Tay scandal.
“If you want to create a chatbot or an LLM that reflects your dataset, it will reflect your dataset – the good, the bad, and the ugly – and it’s important to know that you've created a dataset that is reflecting your values.”
AI and cyber security
The promise and truth of the AI security revolution
Roese also notes LLMs work best with unstructured data, as neural networks seek to create connections of their own rather than relying on arbitrary structures. As a result, he says, businesses need to ensure that their data is “sitting in the right place” to be used for training to avoid having to spend lots of time restructuring data down the line.
Are the majority of firms aware of these requirements at present? “No, they're not,” Roese admits. “And that's very disturbing to be perfectly honest.”
AI is a dress rehearsal for quantum
Generative AI has been hailed as one of the most significant technological developments of the century. At Dell Technologies World 2023, CEO Michael Dell compared it to the invention of the internet or PC.
“Everybody's talking about generative AI as if it is the destination – but it isn’t."
In recent months, many have highlighted the risks of generative AI, with analysts calling it an existential threat, and pioneers calling for a temporary development pause.
But Roese recommends businesses use the big upheaval generative AI has triggered as a “learning experience” to better position themselves for future technologies that will disrupt the sector to a far greater extent.
“Everybody's talking about generative AI as if it is the destination – but it isn’t,” Roese stresses, arguing people are so “shell-shocked” by the headline-grabbing technology that they have failed to give proper thought to what comes next. “The answer to that is actually quite simple in my mind, it's quantum,” he continues.
The primary use case for quantum, Roese explains, is clear: quantum machine learning. He notes while generative AI is branded as “disruptive” and sparking fear in some, it’s just the logical progression of existing technologies.
“Imagine if it now ran at five orders of magnitude higher performance. And that, inevitably, is coming.”
Although we don't expect quantum computers to be widely available for many years, cyber criminals are already stealing encrypted data in the hopes of gaining access in the future.
Maintaining cyber security in the face of developments in quantum computing is something that will become increasingly important as we approach the 2030s.
Experts suggest it’s a matter of if, not when, standards such as AES encryption break down, for example. Once encryption is cracked, the security of data will be wholly undermined.
The private sector isn’t alone in the race to quantum, as many nation-states have already announced huge investments aimed at proactive quantum development and adoption.
In the Spring Statement, UK Chancellor of the Exchequer Jeremy Hunt announced a £900 million ($1.1 billion) fund for quantum computing research, as part of a wider £2.5 billion ($3.1 billion) investment program for UK quantum.
Companies will have to navigate this disruption in the near future, Roese warns, and would do well to use the choppy waters of generative AI as a dress rehearsal for weathering the coming storm of quantum computing.
