With 278 sessions across 18 streams, there was a lot of education and training going on across the three and a half days of the GSE UK conference this year. I thought that I’d share some of what I learned while I was there.
The first session I attended on the Tuesday was IBM Champion Henri Kuiper’s session to the AI stream entitled, “AI AI AI What Has Turing Started?”. Henri started by looking at the computers he had owned and how they had developed over the years, and then moved on to mainframe developments. He explained how the Turing test worked, and quoted John McCarthy from the 1950s saying, “Artificial intelligence is the science of making machines do things that would require intelligence if done by humans”. IBM’s Arthur Lee Samuel in 1959 said, “Programming computers to learn from experience should eventually eliminate the need for much of this detailed programming effort”. Henri talked about Eliza, your personal therapy computer, and much more as Artificial Intelligence (AI) developed. Alain Calmerauer in 1972 developed the Prolog programming language. Henri went on to discuss how AIs work and how they can be trained. He discussed deep learning and reinforcement learning, and Generative Large Language Models (GLLMs). He also explained how transfer learning could be used to avoid training a model from scratch and how it helps to improve the model’s performance on the target task (or domain) by leveraging already existing knowledge. And ended up with Gollems (GLLMs).
After that, I watched Elpida Tzortzatos, an IBM Fellow and CTO AI for IBM zSystems, discuss “AI for Business with Trust and Transparency”.
After lunch I went to a security session with Al Saurette from MainTegrity, who was discussing “Early warning of cyber attacks. Ways to stay ahead of the bad guys”. He highlighted how problems can occur even at sites with the best firewalls and access control because the bad guys can get their hands on stolen credential, trusted staff can go rogue, and people just sometimes make mistakes. He explained that hacking is a business. If the bad actors find a way in, they'll leave multiple backdoors so that they and others can get back whenever they want. They may install timebombs in case they get caught at this early stage. They'll compromise backups to prevent recovery of files. They'll take a copy of your data (exfiltration) to sell. And then encrypt your data and send a ransom demand. That, he explained, was why it was so important to always monitor what was going on and quickly identify anything that might damage your data. He called it integrity monitoring and alerting. You could then tell what files were affected, the time interval during which the attack took place, which userid or job was responsible, and check whether that change had been authorized (to avoid any false positives). The people alerted could also check the files line by line using before and after copies of the data. Because the bad actors can encrypt your data. The software needs to identify in the first seconds if unauthorized encryption activity is taking place and suspend the task. If later, it’s found that everything is OK, then the task can carry on from where it left off. If it’s not OK, you’ve just been saved from a mass encryption activity, and you won’t be sent a ransom demand. Al Saurette also described other early warning capabilities available and how important their use was. The other benefit of using the tools available was compliance with PCI DSS, NIST, GDPR and other regulations.
Lastly on the Tuesday, I watched “Jekyll and Hyde of Generative AI”, presented by Venkat Balabhadrapatruni, a Distinguished Engineer with Broadcom MSD. He started by saying that understanding the use case or business need should drive the right approach – whether that's using Artificial Intelligence, machine learning, deep learning, or generative AI. He explained that generative AI is a branch of artificial intelligence (AI) that focuses on creating content, data, or outputs, based on patterns learned from large volumes of training data.
There's a booming and evolving generative AI ecosystem. He suggested that generative AI will transform how organizations work over the next five years. IP, security, privacy, and ethical concerns will drive a vast majority of large Enterprise customers to adopt well-governed on-prem large language models (LLMs). Organizations will need assistance (LLM selection, in-house training, integration, etc) to fully capitalize on Gen AI value proposition. He then ran through a number of uses of AI.
Venkat Balabhadrapatruni suggested that the positive aspects of Gen AI were: creativity and Innovation; efficiency and automation; content summarization; language translation, and medical and scientific advancements. He then moved on to some challenges and ethical concerns, such as: privacy concerns; hallucinations; resource intensive, regulatory challenges; and lack of transparency. Lastly, he listed the potential dangers and misuses, eg: deepfakes and manipulation; intentional misuse; and bias, misinformation, and fairness. The key takeaways from his presentation were to start with the business need, not with the technology; recognize that generative AI is not the only AI; understand the data and algorithms; the onus is on the user to validate the responses from any generative AI technology; and that the responses from generative AI are only as good as the training data and the specificity of the prompt.
I’ll look at some of the sessions from Wednesday next time.
No comments:
Post a Comment