Asking AI about itselft to see what is coming next

Talking to AI via MyDocSafe
June 21, 2023

AI is here. Some of you have tried midjourney to generate images (the image associated with this post is one of them), asked chatgpt to write an assay, or used copilot to generate snippets of computer code.   AI is already impacting some creative professions, education and IT.   How will AI be used by software vendor like us?  The answers may come by asking that question to Chatgpt.  Why would that make sense? Because the answers are a distillation of what people write about it and what AI writes about it.   We are unlikely to find anything unexpected, but let’s not prejudge the results.

First some context

Within the world of Software-as-a-Service, we see AI employed primarily as part of user interface

  • Voice to text interfaces that translate a spoken word into a computer readable query that is then passed on to the query engine for analysis and response, which is then translated again from text to voice and broadcast to the end user (think of Siri, Amazon Alexa or Cortana)
  • Some software vendors are launching AI-driven customer support bots that can answer questions, as long as the underlying answers have been carefully compiled in a thorough library of self-help articles. AI guesses how to answer new queries based on those articles.  This is similar to chatgpt but the answers are based on a proprietary information rather than public internet

 

What questions are best answered by voice or chat enabled AI?

Voice or chat-enabled AI is particularly effective for answering a wide range of questions that require information retrieval, real-time assistance, or interactive dialogue. Here are some examples:

  1. General Knowledge: Questions about historical events, scientific facts, definitions, trivia, and general information.
  2. News and Current Events: Queries related to the latest news updates, sports scores, weather forecasts, stock prices, and other real-time information.
  3. Navigation and Directions: Requests for directions, information about nearby places, traffic updates, public transportation schedules, and travel recommendations.
  4. Language Translation: Queries regarding translations of words, phrases, or sentences between different languages.
  5. Recommendations and Reviews: Seeking suggestions for movies, books, music, restaurants, hotels, or other products based on personal preferences or user reviews.
  6. Calculations and Conversions: Mathematical calculations, unit conversions, currency exchange rates, and measurements.
  7. Personal Assistance: Setting reminders, scheduling appointments, sending messages, making phone calls, and managing to-do lists.
  8. Health Information: Basic health-related questions, symptoms, first aid advice, information about medical conditions, and medication reminders (note that this should not replace professional medical advice).
  9. Entertainment and Leisure: Engaging in interactive conversations, telling jokes, playing trivia games, or providing virtual companionship.
  10. Educational Support: Assisting with homework questions, providing explanations for concepts, offering learning resources, and guiding through tutorials.

It’s important to note that while AI-powered systems are increasingly capable, there may be limitations and cases where human assistance or expertise is necessary, such as complex legal or medical matters.

 

How about asking AI a little more specific question, such as:

What questions are best answered by voice or chat enabled AI which is tied to a proprietary software interface?

Voice or chat-enabled AI systems tied to a proprietary software interface are specifically designed to provide support and assistance within the context of that particular software. Here are some examples of questions that such AI systems can effectively address:

  1. Software Functionality: Queries about how to perform specific tasks or use various features within the software.
  2. Interface Navigation: Assistance with navigating through the different sections, menus, and options of the software.
  3. Troubleshooting and Error Messages: Help in resolving errors, warnings, or issues that occur while using the software.
  4. Customization and Settings: Guidance on how to personalize the software’s settings, preferences, and user interface.
  5. Integration and Compatibility: Information on integrating the software with other applications or systems and ensuring compatibility.
  6. Data Management: Assistance with creating, editing, importing, exporting, or backing up data within the software.
  7. Reporting and Analytics: Support in generating reports, analyzing data, and interpreting statistical or analytical information within the software.
  8. Automation and Workflows: Help in setting up automation processes, creating workflows, and streamlining tasks within the software.
  9. Security and Privacy: Guidance on configuring security settings, managing user access, and safeguarding data within the software.
  10. Upgrades and Updates: Information about software upgrades, new releases, patches, and how to install or update the software.

It’s important to note that AI systems tied to a proprietary software interface are primarily focused on providing assistance within the software’s specific domain and may have limitations when it comes to answering broader, non-software-related questions or offering advice outside of the software’s functionality.

Nothing surprising here.  AI is simply saying ‘a customer support bot’.

But is it really true? Let’s make a mental experiment.

Imagine that you login to MyDocSafe.  The first thing you see is a dashboard. Within a split of a second you see the status of your signatures, status of workflows and other user actions.  If you can’t see what you are looking for, you will probably navigate to the Signatures dashboard and with a few clicks see who sent what document.

Now imagine that instead of doing that, you login, and the first thing you do is either type or say “did John send the NDA to Lucy yesterday?”, and the answer is “yes” or “no”. Or “which John”, or “do you mean Lucy from Widgets inc?”.  You get a little annoyed and reply with noticeably raised voice “there is only one John in my office and yes, Lucy from Widgets”.  Then you hear the reply “I am sorry to upset you but I cannot find any NDAs sent by John to Lucy.  The only document I could find was sent two days ago. It was a Confidentiality Statement  and was signed by Lucy this morning”.

If the AI is “clever”, it should learn something from this exchange.  Synonyms of popular acronyms for starters. Or maybe it already knows them, but provides too much information so you say “just shut up” mid-sentence.

I am currently not convinced.

Answers 8 and 9 above, however, do show most promise.

The promise of talking to your software rather than making clicks and seeing your words being translated into actions in real time sounds appealing.  But it is harder than it seems.   Let’s list alternative user interfaces through which you can achieve something useful, for example, sending a document for electronic signing. Let’s also rank them from ‘ease of use’ perspective, defined by a subjective measure of energy required to achieve the task, from lowest to highest.

  • Mouse and keyboard

    • Using a mouse, look up a document, right click, go through setup, send.

  • Chatting with a bot – if you can type fast and know exactly how to send commands then the process is quick – but if you type like a 5 year old and ask vague questions, the exchange with the bot can take too many iterations. A scenario could look like this:

    • You: “Find me a document template for NDA”
    • Bot – shows the list of templates
    • You: “Select the one from 9the of May”
    • Bot – opens the template
    • You “Populate the template with the information for John Smith
    • Bot – “There is no information”
    • You: “Fine, for address, select 1 High Street, Glasgow, SG3 4DB; for greeting, say, “hello John, as discussed here is an NDA for you to sign.  Or wait, cancel that, instead use the message template”
    • Bot – Which one?
    • You: “What are the options?”
    • Bot – We have 4: NDA greeting, General message, Welcome to our firm and Engagement letter.
    • You: “Fine, that use the first one.”
    • Bot – Ok
    • You: “Ask for the signature on the dotted line. I will countersign first.”
    • Bot – I Sent the document to you and john@smith.com

  • Speaking with a bot – the above exchange could also be done by voice but that, in our view, takes most brain processing power, time, and emotional energy (just think of the number of times you said “shut up” to Siri or Alexa).

Conclusion

We are entering the experimentation phase of finding the most useful use cases for AI that will (1) be accepted and loved by end users and (2) will make commercial sense to vendors.  If you would like to take part in that journey do sent us your ideas, especially about:

  • types of queries you would want answered by a bot rather than a well-designed dashboard
  • types of actions you would rather perform by interacting with a bot rather than using the keyboard and a mouse

Send your answers to customerservices@mydocsafehq.com.  The best answers will be published and implemented as soon as possible.