The world of artificial intelligence moves quickly, with the potential to leave behind companies that fail to adapt. That, along with advice on how to chart a path into AI, was the message from Dell’s General Manager and VP South Africa, Doug Woolley, speaking recently at ITWeb’s AI Summit in Johannesburg.
Woolley told delegates that AI was not a fad, but was quickly becoming a business imperative, adding that the rapid pace of change was disruptive, while the impacts would be wide-reaching.
Citing Dell’s 2024 Innovation Catalyst Study, he said that 31% of South African respondents admit they’re struggling to keep up with the pace of disruption.
“The whole industry is evolving at pace, and that creates a lot of tension within businesses. Most people understand it will have an impact,” he said.
According to the research, 91% of local respondents believe AI and generative AI will significantly transform their industry.
Woolley highlighted some AI-driven changes already being made within Dell. “For the first time in my life, as a salesperson I didn't negotiate my sales target this quarter; it was given to me, generated by an AI bot. It’s the same for all the sales team,” he said.
He also said that a system of ‘hot quotes’ had been introduced to improve productivity – internal salespeople typically issued lots of quotes, but after being run through an AI system those with a 95% probability of success, or higher, were identified, so the sales team’s efforts were targeted to those clients.
Woolley outlined a future where AI would improve productivity more broadly, thanks in part to the plethora of devices being launched that included AI enhancements, such as dedicated NPU chips. But he outlined that there needed to be a ‘planned approach’ of introducing such solutions into an organisation, where pitfalls, such as use of personal data and cybersecurity considerations were factored in.
The data security challenges go beyond devices though, he added. “Lots of organisations will be introducing their own forms of ChatGPT and Copilot, so it’s critical to ensure it’s a corporate environment, where data is protected, to ensure patents and IP aren't going out into the web through LLMs.”
The storage and management of data could also become a barrier to effective adoption of AI for organisations. Data assets were generally not stored in an orderly fashion, Woolley said, with data in different software and siloes across the organisation, and there was also the issue of unstructured data, such as information from internet sources, to be addressed.
“IT departments may say all the data is in a data lake, or in the cloud, but they sometimes forget about the shadow datasets of the organisation, all the legacy datasets in the business.
“If you don't collate the data and don't utilise all the sources, not just LLMs, but unstructured data, you're going to get caught out. It's about harnessing the data and getting the right insights.”
Supply chain ripples
The growth of AI is also set to be impacted by global infrastructure supply chains, especially with potential US controls, expected in May, imposing export restrictions on the volume of GPUs allowed to other nations. Under guidance issued under the previous US administration, South Africa may only be able to receive 100 000 GPUs per year from US suppliers, which could shape the market. This is likely to lead to similar models as have been developed with public cloud, with centralised ‘GPU farms’ running thousands of GPUs and offering ‘GPU-as-a-Service’ for those organisations unable to justify their own AI infrastructure.
“As an organisation you will have to assess how you augment your on-prem infrastructure and cloud, and ensure you're in control of your own destiny,” said Woolley.
For those organisations that were contemplating investment into their own AI infrastructure, Woolley advised that they look to build an AI factory to create the large quantities of tokens (the processing units used in AI machine learning) needed to support their ambitions.
“By building a factory, you’re able to reduce the complexity and effectively get a repetitive model to churn out tokens, so models can be run as many times as necessary. It's critical to run this in a way that’s automated, with no human interface or interaction.”
Dell has developed an AI factory model with Nvidia, which oversees the hardware, middleware and management stacks. The vendor has also developed reference architectures that can be made available to clients.
For those companies that are ready to move from the AI sandbox, Woolley said Dell has created a test infrastructure in Ireland, which has systems running on each of the GPUs available from the major vendors, as well as providing access to the various LLMs on offer. “Customers can come and test, in a secure environment, any use cases without having to spend money by building your own test bed.”
He added that there are plans to offer a local testing infrastructure in South Africa in the coming months.
On the local front, Woolley was effusive about the local partner ecosystem, and said they can assist with data analytics, help build Copilots and LLMs, as well as digital twins. “The skillsets, the data scientists, they sit in South Africa and they sit in our local ecosystem.”
In closing, he added that the key to success in AI was not about the underlying infrastructure. “It's about the use case and the way you utilise your data, which will give the competitive edge.”
Share