Tag Archives: Disambiguation

Amazon’s Alexa vs. Google’s Assistant: Same Questions, Different Answers

Excellent article by  .

Amazon’s Echo and Google’s Home are the two most compelling products in the new smart-speaker market. It’s a fascinating space to watch, for it is of substantial strategic importance to both companies as well as several more that will enter the fray soon. Why is this? Whatever device you outfit your home with will influence many downstream purchasing decisions, from automation hardware to digital media and even to where you order dog food. Because of this strategic importance, the leading players are investing vast amounts of money to make their product the market leader.

These devices have a broad range of functionality, most of which is not discussed in this article. As such, it is a review not of the devices overall, but rather simply their function as answer engines. You can, on a whim, ask them almost any question and they will try to answer it. I have both devices on my desk, and almost immediately I noticed something very puzzling: They often give different answers to the same questions. Not opinion questions, you understand, but factual questions, the kinds of things you would expect them to be in full agreement on, such as the number of seconds in a year.

How can this be? Assuming they correctly understand the words in the question, how can they give different answers to the same straightforward questions? Upon inspection, it turns out there are ten reasons, each of which reveals an inherent limitation of artificial intelligence as we currently know it…


Addendum to the Article:

As someone who has worked with Artificial Intelligence in some shape or form for the last 20 years, I’d like to throw in my commentary on the article.

  1. Human Utterances and their Correlation to Goal / Intent Recognition.  There are innumerable ways to ask for something you want.  The ‘ask’ is a ‘human utterance’ which should trigger the ‘goal / intent’ of what knowledge the person is requesting.  AI Chat Bots, digital agents, have a table of these utterances which all roll up to a single goal.  Hundreds of utterances may be supplied per goal.  In fact, Amazon has a service, Mechanical Turk, the Artificial Artificial Intelligence, which you may “Ask workers to complete HITs – Human Intelligence Tasks – and get results using Mechanical Turk”.   They boast access to a global, on-demand, 24 x 7 workforce to get thousands of HITs completed in minutes.  There are also ways in which the AI Digital Agent may ‘rephrase’ what the AI considers utterances that are closely related.  Companies like IBM look toward human recognition, accuracy of comprehension as 95% of the words in a given conversation.  On March 7, IBM announced it had become the first to hone in on that benchmark, having achieved a 5.5% error rate.
  2. Algorithmic ‘weighted’ Selection verses Curated Content.   It makes sense based on how these two companies ‘grew up’, that Amazon relies on their curated content acquisitions such as Evi,  a technology company which specialises in knowledge base and semantic search engine software. Its first product was an answer engine that aimed to directly answer questions on any subject posed in plain English text, which is accomplished using a database of discrete facts.   “Google, on the other hand, pulls many of its answers straight from the web. In fact, you know how sometimes you do a search in Google and the answer comes up in snippet form at the top of the results? Well, often Google Assistant simply reads those answers.”  Truncated answers equate to incorrect answers.
  3. Instead of a direct Q&A style approach, where a human utterance, question, triggers an intent/goal , a process by which ‘clarifying questions‘ maybe asked by the AI digital agent.  A dialog workflow may disambiguate the goal by narrowing down what the user is looking for.  This disambiguation process is a part of common technique in human interaction, and is represented in a workflow diagram with logic decision paths. It seems this technique may require human guidance, and prone to bias, error and additional overhead for content curation.
  4. Who are the content curators for knowledge, providing ‘factual’ answers, and/or opinions?  Are curators ‘self proclaimed’ Subject Matter Experts (SMEs), people entitled with degrees in History?  or IT / business analysts making the content decisions?
  5. Questions requesting opinionated information may vary greatly between AI platform, and between questions within the same AI knowledge base.  Opinions may offend, be intentionally biased, sour the AI / human experience.

Intent Recognition: AI Digital Agents’ Best Ways to Interpret User Goals

Goal / Intent recognition may be the most difficult aspect of the AI Digital Agent’s workload, and not Natural language processing (NLP) or Voice Recognition.

Challenges of the Digital Agent
  • Many goals with very similar human utterance / syntax exist.
  • Just like with humans trying to interpret human utterances, many possibilities exist, and misinterpretation occurs.
  • Meeting someone for the first time, without historical context places additional burden on the interpreter of the intent.
  • There are innumerable opportunities to ask the same question, to request information, all achieving a similar, or the same goal.
Opportunities for Goal / Intent Accuracy
  • Business Process Workflows  may enable a very broad ‘category’ of subject matter to be disambiguated as the user traverses the workflow.  The intended goal may be derived from asking ‘narrowing’ questions, until the ‘goal’ is reached, or the user ‘falls out’ of the workflow.
  • Methodologies such as leveraging Regex to interpret utterances are difficult to create and maintain.
  • Utterances are still a necessity, their structure, and correlation to Business Process Workflows.  However, as the knowledge base grows, so does the complexity of curation of the content.   A librarian, or Content Curator may be required to integrate new information, deprecate stale content, and update workflows.
Ongoing, Partnership between Digital Agent and Human
  • Business Process Workflows may be initially designed and implemented by Subject Matter Experts (SMEs).  However, the SMEs might not have predicted all possible valid variations of the workflow, and achieve a different outcome for the triggered goal.
  • As the user traverses a workflow, they may encounter a limiting boundary, such as a Boolean question, which should have more than two options.  Some digital assistants may enable a user to walk on an alternate path by leveraging ‘human assisted’ goal achievement, such as escalation of a chat.  The ‘human assisted’ path may now have a third option, and this new option may be added to the Business Process Workflow for future use.