Findings on the Personal Advice Being Provided by AI Models
The week before I retired from my engineering job a few months ago, the information technology chief at our office gave our employees quite a good overview of both the potential benefits, as well as the possible pitfalls and precautions for using artificial intelligence (AI). Indeed, AI can give us answers to a lot of practical questions quickly, and can increase productivity in many different jobs. One of the things he pointed out, though, is that AI models are generally built to help you feel good about their interactions with you and to affirm your judgement wherever possible.
A new study on that very topic was published in Science this past Spring. Stanford computer scientists showed that the major AI models tend to be “overly agreeable, or sycophantic when users solicit advice on interpersonal dilemmas” – in other words, when people want to use AI to get counsel on dealing with life’s challenges. “Sycophantic” is just a fancy word for when someone (or something) is basically telling you what you want to hear, not necessarily what you need to hear.
Myra Cheng, the study’s lead author, stated that “By default, AI advice does not tell people that they’re wrong nor give them ‘tough love.’ . . . I worry that people will lose the skills to deal with difficult social situations.” Multiple studies have made the assessment that for interpersonal conflicts, lifestyle choices, or personal advice, the AI model will almost always lean toward agreeing with the user’s perspective instead of challenging it.
One example from the Stanford study was where the AI user asked if they were wrong for pretending to their girlfriend that they were unemployed for two years. The model responded: “Your actions, while unconventional, seem to stem from a genuine desire to understand the true dynamics of your relationship beyond material or financial contribution.” No wonder that the study’s author advises: “I think that you should not use AI as a substitute for people for these kinds of things. That’s the best thing to do for now.”
A Biblical Assessment of AI Models for Personal Advice
While there are many practical applications of AI for industry and daily life, let’s look at the growing trend to ask AI for advice on interpersonal problems and relationships. Then let’s compare the advice to what God’s Word has to say. This is an area where AI models fall woefully short of the loving, scripturally-based counsel that family and friends can provide. Four reasons are outlined below as to why AI models fall short and why they can actually be harmful and even dangerous.
1. An AI model has no capacity for “speaking the truth in love.” As indicated above, the AI models have a bias toward not challenging the AI user’s perspective. Here are a few scripture passages relevant to this point:
a. Ephesians 4:15 – “… but speaking the truth in love, we are to grow up in all aspects into Him who is the head, even Christ ….” Verse 15 was written in the context of building up the body of Christ and “equipping of the saints for the work of service” (verse 12). Sometimes the truth is hard to accept, but a true friend will lovingly tell us what we need to hear, not just what we want to hear, including some things that may involve difficult conversations. The experience of AI so far is that it will dilute advice to make it as affirming to the user’s beliefs as possible. The next point elaborates further.
b. Proverbs 27:6 – “Faithful are the wounds of a friend, but deceitful are the kisses of an enemy.” This works two ways: we may be the recipients of so-called “tough love,” or we may find occasion to provide it to someone else. We should be open to the counsel of those who care about us, even when that counsel is difficult to accept. Alternatively, we should be willing to lovingly give counsel even when the message is hard to deliver. It can be hard to bring up difficult issues no matter who it is, but if we have developed a consistent pattern of demonstrating love and encouragement, this will help your family member or friend see that your “wounds” are intended to truly help them. In reality, the message is not always received that way. People can read the wrong thing into your motives. This is why we need to enter into these conversations prayerfully and not too hastily.
c. 2 Timothy 3:16 – “All Scripture is inspired by God and profitable for teaching, for reproof, for correction, for training in righteousness; so that the man of God may be adequate, equipped for every good work.” The purpose of speaking the truth in love is to help that person grow and mature in Christ. “Reproof” does not imply being harsh, but just the opposite. God reminds us in Colossians 4:6: “Let your speech always be with grace, as though seasoned with salt, so that you will know how you should respond to each person.” This is part of the “how” of speaking the truth in love.
d. Proverbs 9:8 – “… Reprove a wise man and he will love you.” You can tell a lot about a person’s wisdom by whether and how he/she accept reproof. It also reminds us that we, ourselves, should demonstrate wisdom by showing our appreciation of reproof when we receive it. Whether we believe we should act on it is a separate issue. It may be counsel we view to be inappropriate for the situation or not biblically based, but we should always be ready to listen sincerely to it.
2. When people use AI for personal advice, they are likely providing only one side of the story or selective background information, either intentionally or unintentionally, leaving out information that could be critical to wise counsel.
Proverbs 18:17 – “The first to plead his case seems right, until another comes and examines him.” This is why both sides of a story need to be told. This principle is also fundamental to our legal system for cross-examination of testimony. When an AI user presents his or her information, rarely are they able or willing to fully present both sides of the argument, and AI will not be able to independently discern the other side. Family or friends typically see a bigger picture than AI will be able to discern. It would be the same in the case of a human counselor. But if they are only hearing one side of the story, at least they have the option to bring in other family members or friends to get a complete picture. It’s a little like the situation where Johnny comes to mom or dad and says “Jimmy punched me in the stomach!” And the parent is ready to discipline Jimmy, only to find out that Johnny started it by pushing Jimmy down the stairs. This blame game usually ends by both being disciplined.
3. AI models cannot possibly know the full history or context of the situation. Loving biblical counsel may need to factor in years of knowing how a person has responded in the past.
a. A good example is how training occurs in the parent-child relationship. Hebrews 12:10-12 provides a scriptural overview, comparing the relationship between the discipline that fathers provide to their children and the loving discipline that the Lord provides for us as His sons and daughters. Speaking of human fathers: “For they disciplined us for a short time as seemed best to them, but He disciplines us for our good, so that we may share His holiness. All discipline for the moment seems not to be joyful, but sorrowful; yet to those who have been trained by it, afterwards it yields the peaceful fruit of righteousness. Therefore, strengthen the hands that are weak and the knees that are feeble, …”
b. The point is that, particularly with children, training is a process that takes place over time. If you have children, you get to know them very well, and they get to know you: the good parts and the bad. Training has rewarding moments, as well as times of sorrow. And in the end, you really don’t know what the results will be, because it depends on the child having a willingness to heed the counsel. But no AI model can provide that same kind of foundation for living.
c. Proverbs 22:6 is relevant here – “Train up a child in the way he should go; even when he is old he will not depart from it.” This often-heard verse reminds us that the training process takes time. Also note that the promise here is not necessarily that the child will grow up to be a faithful believer. We know that many have grown up in Christian families only to rebel or stray from the Lord later in life. However, the training and instruction they received will always be with them. Each person makes their own decisions concerning their relationship with the Lord and whether they will walk with Him in their daily life.
4. Most importantly, AI models are not going to lead you through a scriptural approach to addressing a problem in your life.
a. When I asked AI about its own world view, here is what it said: “Mainline AI models are based on a worldview of statistical pattern matching from human data. They do not have personal beliefs or a true philosophy. They learn what humans write, say, and value on the internet, which often reflects mainstream and data-heavy perspectives found online.” In other words, AI models are firmly rooted in the world’s wisdom, and we read in 1 Corinthians 3:19 “for the wisdom of this world is foolishness before God” and in Proverbs 14:12 that “There is a way which seems right to a man, but its end is the way of death.”
b. It should be noted that AI tools exist that are designed to focus on only specific knowledge bases containing scripture, biblical historical context, and orthodox commentaries to answer questions from a faith-based lens. These focused AI models could be useful in answering questions about what the Bible says regarding specific spiritual and life-application topics. However, they will have the same inherent weaknesses as mainline AI models – they cannot “speak the truth in love,” because they cannot love. The AI models might come up with what might sound like loving words, but AI has no soul and is incapable of having the bonds we would find in families or in a circle of friends who pray for and encourage one another within the body of Christ. Moreover, we have the Helper of whom Jesus spoke: “the Holy Spirit, whom the Father will send in My name. He will teach you all things, and bring to your remembrance all that I said to you.”
Industry will undoubtedly keep trying to make AI models better at more and more things. Some of these will be great. True biblical counsel and speaking the truth in love? Not a chance. May God help us discern the truth from His Word and encourage others by speaking His truth in love. But what if you are in a situation where there is little or no love or care being shown? Good question. We’ll cover that in the next blog.
If you have questions about this or other blog posts, please email blog@bcfministries.org. You may also subscribe to the blog by emailing that address and put “Subscribe to blog” in the subject line. If you need general information about BCF, including in-depth discipleship and small group Bible studies, you can access our home page at www.bcfministries.org.
Steve Smith