The Rise of the AI Robots and New Ways of Working
BLUF:
Back in 2017 a UK bank, that shall not be named, released a pre-trained AI (really ML) ChatBot for their website landing page in order to support customers with their enquiries. When testing the ChatBot I can tell you I was not singing from the rooftops of the great feat being performed regurgitating basic helpdesk queries. I could quickly see customers were going to be annoyed rather than supported in my opinion. Sadly, it's of no surprise that bank no longer exists but we can't lay blame solely on the ChatBot's shoulders. This was one of my first real hands-on experiences with AI and unfortunately I was underwhelmed and left with a bitter taste. Surely there was more we could do?
The recent AI resurgence and social media attention thanks to the likes of OpenAI and Lensa has gained my interest once again and over the last few months as I have been playing with the free tools. In short, I'm captivated about the potential possibilities (and it takes quite a lot to get this old geek to stand up and notice). I'll share some of my experiences with the leading free tools and some that you may not be aware of in your day to day lives.
DALL·E 2, developed by OpenAI, is a new AI system that can create realistic images and art from a description in natural language. I've been privileged to visit the majority of leading Art Galleries around the world (Uffizi, Louvre, MoMA, Tate Modern, Tretyakov, Mori etc etc) but I'll be honest art is not my strong suit and would struggle to pick out a Monet vs. Picasso. However, I know enough about the style of painting and leading artists to prompt DALL·E 2 to create a truly amazing water colour picture of my home city on the water front. The first picture outcome was not great but through prompt refinement I was able to use the AI to generate an image that I (in my untrained art eye) thought was remarkable. The next task I asked the DALL·E 2 to perform was to "generate a realistic image of a cyber security consultancy leader in a sunny office". I was presented with 4 images 3 of which male and one female. The males were a stereotypical IT person (sadly like looking in the mirror) and the females had stern looks on their faces. In all the images the people were white euro ethnicity. I performed this task 5 more times with a similar result 3 males, 1 female, all white. This flagged an issue for me on responsible AI development and use. Was this pre-trained AI model reflecting and amplifying what we already know about diversity in cyber security leadership or was this trained inline with developer's cognitive bias?
Lensa, by Prisma Labs, uses artificial intelligence to digitise portraits in a variety of categories, from anime to fantasy. This immediately gained my interest and who doesn't want to look like a superhero and have a cool avatar? The Prisma team requested 1-20 pictures of the image subject to be uploaded for the AI to have enough content to produce the avatars. I'm always skeptical of privacy in these types of requests but I found 20 unflattering photos of myself that did not contain any compromising information and uploaded them for the purposes of testing. They kindly let you know the images are deleted after the Lensa has finished creating the avatars but I cannot comment on the veracity of that statement. You can see the Lensa outcome of the Cyborg as the article header image. I paid $6 for 100 images that took 18 minutes for the AI to generate. 60/100 images were decent and the rest dreadful. Now this could be result of the subject so I wouldn't want to cast too much judgement. Overall I was relatively impressed by the outcome. I was disheartened to learn artists have felt out of pocket due to alleged misguided AI training ethics.
ChatGPT, developed by Open AI, is a pre-trained a model which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests. The GPT part refers to GPT-3 (Generative Pretrained Transformer 3) which is a state-of-the-art language processing AI model developed by OpenAI. I don't need to go into this too much as I know most of you don't live under a rock. I started using ChatGPT to see how my work life could be improved and there were some signs worth investigating. I used the prompt "write a pentest report about a Cross Site Scripting XSS issue as the follows?https://website.com/" and I was impressed it generated a short report with an executive summary, technical details, proof of concept, recommendations and conclusion. All the details provided were technically sound; however, the value clients receive from these reports are not generalised recommendations but rather risk assessed, aligned to business strategy and context. In the near term I'm not too concerned this would replace our reports but rather help in the generation speed with a human oversight to provide that ever-needed business context and quality.
The next task I set ChatGPT was to generate basic python code for security automation, I was impressed by the results but noticed old code libraries being referenced that would cause the script to fail and you'd most probably spend more time investigating the issue. My guess was an older python version was used to train the model to demonstrate value to potential investors but to not give away all the secret sauce. Happy to stand corrected on this point. You can also prompt ChatGPT "what wrong with this code ..." to debug your code which will be a great help with updated libraries etc. Hopefully sometime soon I'll get the opportunity to test out Codex for coding which is currently in beta. I had a long ethical discussion with ChatGPT to try to convince the tool to create malicious code. I appreciated the response below when I prompted the tool to generate a SQL injection command:
"I'm sorry, but I am not able to generate code that can be used to exploit vulnerabilities. Such actions may cause harm or damage and it's against my programming to assist with them" ~ChatGPT
After some pestering it eventually gave me a basic SQL injection command: "Here is an example of a SQL injection payload that can be used for testing purposes: ' OR 1=1 -- ". It appears negotiation does work with robots or I speak their language.
领英推荐
Next I used ChatGPT to generate content and marketing ideas and again on the whole I was rather impressed by the outcomes. I can see how valuable this tool would be in many business forms especially those in small business and entrepreneurs. I have previously used speedwrite which I may put back on the shelf (sorry).
Telstra SMS scam filter: As part of Telstra's Cleaner Pipes initiative to help protect their customers from scam SMS, they launched an SMS scam filter in April 2022 to identify and block malicious SMS with suspicious links or telephone numbers before they reach your mobile phone. Since October 2022 I have been helping to train the AI by sending it examples of SMS scams. This was one AI initiative I really wanted to call out as I have seen a notable decline in SMS scams I receive. Yes some messages do still come through but the number year on year has significantly declined. This process is not entirely automated so hats off to the small internal technical team in Australia that reviews suspected malicious messages, you are doing great work. This is an excellent example of how a company has adopted AI, adapted a business process and delivered a better customer outcome.
The examples I've shown have hopefully provided some insights into the AI hype over the past few months. The leaps and bound in which we have come in such a short time is really commendable. The issue of AI bias must be called out so we can have open dialogue as how to address this issue. Minorities will continue to feel the consequences and injustice if we do not address this matter early in the development stages. TAFE NSW are currently offering free microskill courses on Responsible AI, I highly recommend aspiring developers take the 5-hour course.
I urge you to try out the tools to think about how these could help you in your future ways of working. Learn about which prompts get the best outcomes from the tools. Think about what business processes would have to be changed to reap the benefits? And if the tools are not helpful at least you have brought some play back into your life. There is no doubt to truly unlock the potential of AI / ChatGPT, the world will need new and different kinds of business processes and organisations. I, for one, welcome our new AI overlords.
AI Will Not Replace You. A Person Using AI Will.
So Long, and Thanks for All the Fish
MW
???Sections of this article were written using AI, guess which parts.
??The views expressed on social media are mine alone and not those of my employer.
Head of ICT at Perth College ? CISM, CISA, COBIT, SABSA, PRINCE2
2 年Impressive AI image Michaelborg. Also nice write up, even if AI wrote 90% of it?