To connect effectively with AI tools as a human, asking the right questions is key. This is time taking but essential for a valuable use of AI tooling. Remember that input data is the source of all answers. So, break-through inventions or insights cannot be expected from AI yet.
Also the objectivity of AI tools should be questioned since the algorithms are written by humans. Be cautious and always wonder when working with AI tools, as this article shows.
Understand, the fast majority of source (input) data any AI tool uses, has been created by humans. When AI publishes through humans, “new” information, it is only based on what is already available (the input data) and is therefore never new(s). Unless AI works in a laboratory or conducts interviews and is able to work statistically or philosophically, it is dependent on collaboration with humans.
The 5 topics tested
The testing or communicating entailed 5 topics to find out how intelligent the algorithms are, and how it “behaves”. For every topic I started a new chat.
Note: the topics are my fields of expertise.
The ChatGPT topics were discussed in this exact sequence.
1. My Personal information
It refuses to show anything about me as a person, address etc., due to privacy. Fair enough. But is also refuses to share information when I give my exact LinkedIN name and position which is public information.
It does not even want to show the general LinkedIN website.
2 History: which lessons did we learn from WWII?
When you push the regenerate button, it generates new Lessons Shown as 2/2 but also chooses to reuse some.
One lesson is about international cooperation. Well after WW I the League of Nations was erected with its principal mission to maintain world peace, followed by WW II. And given the number of conflicts/wars after WW II, the newly erected United Nations was also not a huge success.
There were 5 lessons in both answers but (Recognising) Propaganda was not mentioned. To my (40 years historical) knowledge, propaganda was key to start WWII supported by the people. Open for discussion.
So I asked about Propaganda.
To understand how ChatGPT comes to these Lessons, a follow up question.
Useful summary however my criteria would focus more on understanding the causes leading to war and its prevention, hence my suggestion to include - Recognise propaganda - as a key lesson learned.
The final selection of most important Lessons Learned from WWII is for human mankind to decide.
3. Geopolitics: 9-11 the Twin Towers collapse and the controversy
There is controversy about the people and organisations behind the Twin Towers attack in New York, Sept 11th, 2001. To not confuse ChatGPT, two separate chats were created.
3.1 Who was responsible for the attack on the Twin Towers in New York September 11th, 2001?
ChatGPT tells the general narrative pointing at 19 Middle East hijackers, but contradicts itself when asked about the airplane crashing in the Pentagon.
ChatGPT speaks of a heavily guarded Pentagon but at the time of the crash everyone was on high alert and still nothing was done to prevent an approaching airplane from crashing into it. Also do the security cameras only show fragments of the airline crash in the Pentagon. Last, this attack demanded trained pilots who had experience flying commercial airliners, which the terrorist did not have …
Or ChatGPT presents (the input data) wrong.
Even more interesting when ChatGPT apologises for not providing working internet links (first paragraph) but providing them anyway (second paragraph) in the same answer.
Asking for a LinkedIN link is refused (See topic 1) but for some algorithmic reason discussing a controversial matter result in working internet links.
When asking why it provides links while also saying it is not able, ChatGPT contradicts itself again.
3.2 Which American government agencies were responsible for the attack on the Twin Towers in New York September 11th, 2001?
The answer speaks of lack of evidence and speaks of conspiracy theories, mentioning the 9-11 National Commission who concluded that terrorists were responsible for the attacks on the Twin Towers.
4. Business: Input regarding our low code solution.
We developed a low code solution for managing data. By trial and error, we discovered the market segment which needed our solution. ChatGPT might give new insights and this is why I asked open questions. I started asking about a low code backend in general to get ChatGPT on the right page, as you do with a human.
Answers:
ChatGPT mixed up a low code application & data management solutions in the first answer. This resulted also in the wrong market segments. I asked for a new answer and limited the scope to low code data management solution. This time it provided several (not one!) relevant market segments which is an important insight for us. A provided strategy was kind of standard but very short and clear described.
Finally a asked for a slogan and provided the solution’s name and its Unique Selling Points (USPs).
Topic 5. Academic research: if a Power Model exists and related questions.
in 2008 I created a Power Model called Dunamis which is not published yet, so this info is not included in ChatGPT’s input data. Curious to see if something exists and if ChatGPT can support in creating a power model. We start with introduction questions.
Note:
A previous Google search for existing power models resulted in billions results, but nothing relevant. Qwant (French based search engine) shows better results.
ChatGPT does not give words for the axes but describes 5 contexts which are represented in the Dunamis Power Model. This feedback adds to the understanding of power dynamics in general.
The second paragraphs mentions institutions to be placed in the Dunamis Power Model, to be published later this year. The final paragraph describes the power dynamics and evolvement of the holders (institutions), resulting in their repositioning and size in the Dunamis quadrants as their power increases or decreases over time.
Since the WHO is mentioned, and not a financial institution e.g., Central Bank, I ask for a further explanation.
The next question for positioning the WHO in the Power Model, is it Democratic or not?
Follow the Money or “Money talks, bullshit walks”.
The last question to provide background information about power and power models presented useful information but again, no direct internet links.
Conclusions based on the topics discussed
· It formulates the answers clear and concise, important for the value it delivers
· As a superficial discussion partner it works fine given the new insights and valuable summaries
· It, or the input data or its processing, has odd inconsistencies, see topic 3
· Business and academic research can greatly benefit from using ChatGPT
· Remarkable feat that ChatGPT never shows (refuses to show) direct internet links except when challenging a mainstream opinion on a controversial issue. This raises the questions why, and which other overarching commands are operating within this AI tool.
Fredrik the Frisian
Comments