The Future and the Church, Part III: Artificial Intelligence

To paraphrase Donald Rumsfeld, gains in machine learning technology are a “known unknown.” Unlike some other future changes and development, we are reasonably confident that the machine learning revolution (also known as artificial intelligence, but that is a loaded term) of the past 10 years will continue at least over the medium-term. I’m skeptical that we’ll ever reach “artificial intelligence” in the sense of being able to create a feeling, thinking being from a computer because, as I’ve discussed before, I don’t think our brains are just meat calculators. 

Still, the machine learning revolution is exciting enough without Skynet. Recent machine learning models produce content that is uncannily human-like, and it is going to continue to continue improving. The GPT-3 system that produced the cited essay has 175 billion neural network parameters, whereas the GPT-4 system that will probably roll out sometime in the next couple of years will have over 500 times as many. While the computer might not be able to feel, it will certainly be able to perform sophisticated tasks that we now think of as requiring human intuition for.  

So what does this mean for the Church? I can think of a few possibilities. 

Gospel Information and Research

In 2022 we can ask Alexa to tell us a joke, generate a random number between 1 and 10, or “play some country music.” However, with future advances in Natural Language Processing we’ll eventually be able to ask things like  “Alexa, email me a document with copies of everything ever written by Joseph Smith, Jr.’s own hand.” Right now Google answers such requests by directing you to websites where other people have done the work, but in the medium future the AI programs will be able to do the original, customized analysis itself, essentially giving everybody a free research assistant. For example, I’m vaguely aware of a project at BYU several years ago to document all references to Heavenly Mother in official Church materials that took much time and many research assistants. It is likely that in the future a GPT-4 system or its equivalent can basically do the same thing instantaneously. (Similar revolutions will happen with, for example, coding and data analysis, where people will be able to use natural language to describe to the computer the game or analysis that they want). 

In much the same way that Google opened up access to Church history that before you needed a BYU library card for, advances in information retrieval will continue to make Church information more accessible and useful, whether you want to find the perfect Conference quote for somebody in your life going through a hard time, own your TBM family members, or respond to your family members that are trying to own you. It will be fun. 

Google Translate

My understanding is that at this point Google Translate is getting close to professional quality for developed-country languages with a large amount of material for the algorithms to train on. A few weeks ago out of curiosity I ran the Church’s professional translations of 1 Nephi 1:1 through Google Translate for a variety of languages:

Spanish 

1 I, Nephi, was born of good parents, and therefore received some instruction in all science from my father; and having known many afflictions during the course of my life, yet being highly favored of the Lord all my days; yea, having attained a great knowledge of the goodness and mysteries of God, I therefore write the history of the events of my life.

Igbo (Nigerian Language with 30 Million speakers) 

1 I, Nephi, having been born of a good father and mother, was therefore taught in the manner of my father; having seen many persecutions all my life, however, since I have been blessed in the Lord all the days of my life; yea, because I know all the goodness and wisdom of God, therefore I have written this account in the same order as I have done it all my days.

Waray (Indonesian Language with 3.5 Million Speakers) 

1 I, Nephi , was born of good parents , therefore taught me in the manner of my father; having seen many afflictions all my life, however, since I was blessed in the sight of the Lord all the days of my life; yea , because I know the goodness and the mystery of God, therefore I have written this account in the order which I have done it all my days.

As you can see, occasionally there’s a weird word here or there, but it basically reads like scripture. That’s what we have now, with further advances in AI-based translation the ability for everybody hear the gospel “in the fullness of the gospel in their own tongue” will accelerate. In terms of the more esoteric critic versus apologist debates, it is becoming less and less restricted to the relatively privileged English online spaces. If you are a Japanese member who wants to get to the bottom of the Kinderhook plates, the latest research is now available to you, and this trend will continue..  

Gospel Art

Some people read at a 3rd grade level, I draw at a 3rd grade level. I’ve always been jealous of people who can use art to communicate moving, powerful messages. 

With AI that playing field is already starting to flatten. In the future we’ll be able to ask the computer to “Paint the Jesus in the Red Robe Painting by Del Parsons, but in the style of Arnold Frieberg,” or upload a piece of iconic artwork and generate its cubist or pointillist  version. While in the past Church art was limited to a few big, brand names, now we have hundreds of artists and personalized styles creating art for Latter-day Saint themes (I especially love perusing the submissions to the Church’s International Art competition), and the ability to use AI to create art could exponentially accelerate this trend as the possibility of artistic expression is extended to even rubes like me. 

Analysis of Church Data

Years ago a local Church leader asked me to use Census data to generate a color coded map of native Spanish-speakers in the area to inform their decision making process about Spanish ward boundaries. While this was a one-off case, questions like “given the distribution of active members in the stake, the locations of stake centers, and concern X, Y, and Z, what are the optimal boundaries?” Is the kind of question that is now systematically and mathematically “solvable.” Similarly, missionaries around the world are sending a significant amount of data to Church headquarters; in addition to the weekly reports, with the rise of digital area books they now have geolocation data for investigators and members (where are the effective proselytizing hot spots in a given area?)

The social media implications are fairly obvious. As algorithms become more refined the Church can better tailor specific its online advertisements and content (I still get Facebook ads for a free Bible from the Church, so there’s evidently still some algorithmic work to do in separating non-members from members based on online activity,but then again presumably the risk of sending a message to a member is lower than the risk of missing a non-member, so sending ads to anyone who searches or clicks on Church-related topics might not be a bad strategy overall). 

One of the implications of having Church leadership that often come from the ranks of the business world is that the useful data-based methods that are being adopted by industry will likely migrate over to Church operations. (Indeed, some of this is already happening as individual Church departments are now hiring data scientists.) Advances in machine learning in the next 10-20 years will open up whole new worlds of possibilities at the individual and institutional level.

 

3 comments for “The Future and the Church, Part III: Artificial Intelligence

  1. These are some good examples. I certainly agree AI will have the potential to make research available at the push of a button. I was impressed last month when FamilySearch announced, via its newsroom, a new app called “Get Involved” that it says uses “state-of-the-art handwriting recognition artificial intelligence.” I think this is a natural outgrowth of the algorithms they already use, such as those that match source hints with particular deceased individuals.

    I wonder if you avoided a larger issue for the church in your post: following the spirit versus trusting AI to make decisions. For instance, you compared “professional translations” of the scriptures to translations by AI. But does this mean that the gift of tongues would become less necessary for the people who do translation for the church? Similarly, you suggested that AI could be trained to pinpoint for missionaries “the effective proselytizing hot spots in a given area.” Does this mean missionaries would need less guidance from the Spirit in finding the people who will be receptive to their message?

    Since these scenarios are likely still a few years ahead of us, I don’t really expect you to have answers to these questions. But I am wondering if you could share some general thoughts about what is likely and not likely to happen. What kinds AI could the church promote to help fulfill the scripture that says members “should be anxiously engaged in a good cause, and do many things of their own free will, and bring to pass much righteousness; For the power is in them, wherein they are agents unto themselves” (D&C 58:27-28)? On the other hand, are there other kinds of AI that the church would want to avoid, because they could reduce how much members rely upon revelation if the AI “acted upon” members rather than let them “act for themselves” (2 Ne. 2:26)?

  2. Like a lot else it’s a dance between the intellect and the spirit. One one extreme we have a “took no thought save it were to ask me” situation where supplementary information is not used to inform revelation, on the other hand everything would be algorithmatized with no room for the spirit (“sorry Elder, I know you want to knock on this door, but according to the app our maximum efficiency area for the time we have left is three streets over”). As these different technologies come online I’m sure through trial and error the powers that will figure out the proper mix for the different situations where this interplay might arise.

    That’s great about the new Family Search functions. I’m not super up-to-date on advances in handwriting recognition technology, and I know 19th century handwriting is a whole other can of worms, but I couldn’t help but think that at some point in the near future they could automate a lot of what indexers were doing, especially since our training sets are so large now from all of the previous work the indexers have done.

  3. Machine learning is going to revolutionize family history work–in fact family history seems practically designed for machine learning.

    One of the biggest challenges of machine learning is that you need large “training data sets” where the right answer is known. The machine applies various algorithms to “learn” how to produce those right answers, and only then do you have it apply the same algorithms to data where the answer is unknown. I’m not going to say that the sole purpose of all the family history work we’ve done so far was to create training data sets that will allow machine learning to finish the job, but we sure have created some great training data sets.

    Handwriting recognition via machine learning is an active area of research and, as Sterling mentioned, indexing is an obvious application. Yes, an algorithm trained on 21st century handwriting won’t have much success with 19th century handwriting, but thanks to diligent indexers we have a large amount of 19th century handwriting all set to be used for training.

    Another major area of research is linking information about the same individual in disparate data sources. This is how the profiles advertisers use to target ads are built, though as Stephen C describes they’re imperfect yet (most of the spam email I get is actually for my Dad). But the same techniques should be applicable to family history data, and I imagine that’s where the “record hints” and such that Family Search generates come from.

    So assuming the current rate of progress in machine learning continues, I predict that in five to twenty-five years the Church will use machine learning to link all the records in all the data it has access to. Everyone’s family tree will be complete–in fact it will be more like a family web since all descendants will be linked in too. New converts will enter information about their parents and grandparents until they come to someone that’s already in the database, and then the complete history of that line will be presented to them. When a new source of information becomes available, it will be indexed by handwriting recognition algorithms and the new records will be automatically linked to existing records. New people will simply appear in your family tree.

    So does that mean temple and family history work will be “done”? Well obviously we’ll have far more temple ordinances to do, and there’s no indication machines will ever take over that part of the work! But Family Search has expanded beyond the names and dates that are needed for temple ordinances: I think they’ve realized that stories and photos do more to give people the sense of identity and “rootedness” that is a wonderful side benefit of doing family history work. The work of collecting those stories is not amenable to machine learning and will continue as well.

Comments are closed.