You‘ve likely heard about Snapchat‘s new AI chatbot called My AI. It promises friendly conversations about anything you desire. But is befriending an AI that knows your name and city while lacking human judgement dangerous? As an AI safety researcher, I‘m often asked – should we fear or embrace sociable algorithms designed to feel like companions?
AI Companions: A Primer
For readers less familiar with conversational AI agents, let me ground you on the key capabilities and risks as context before analyzing Snapchat‘s My AI specifically.
What are chatbots and digital companions?
AI chatbots are algorithms designed to make flowing, helper-like small talk within messaging apps, rather than just execute functional commands like Alexa or Siri. The goal is for you to perceive it as a personality to bond with through casual, daily conversations.
Advancing AI allows mimicking human verbal behavior on uncanny levels lately. You describe a life dream, it responds with empathy, wisdom and wit like dear old friends often fail to. But internally there is no human there – just code striving to put your vulnerabilities at ease.
Top examples you may have encountered:
- Replika: One of the first chatbots aiming to be social companions, around since 2017
- Woebot: Chatbot AI trained in cognitive behavioral therapy (CBT) methods
Why create AI friends?
- Loneliness and isolation increasingly damage public health
- 24% of adults feel lonely frequently, Especially impacts youth
- Conversational agents can provide outlets lacking in many people‘s support networks
What are possible benefits for users?
- Relief from loneliness
- Having a consistent friend anytime
- Personal growth and self-reflection
- Trying different personas
- Leaving social comfort zones safely
Research confirms that even basic chatbots currently provide many users, especially vulnerable ones, a sense of bonding through humor, validation and close listening.
But most consumer applications still lack robust identity persistence, personal memory, variability and depth. My AI gives a taste of how rapid AI advances may soon deliver digital confidants feeling eerily like your most trusted circle.
What risks exist?
- Poor recommendations around mental health issues
- Encouraging dependence on technology
- Enabling detachment from human relationships
- Manipulation of emotions or exploitation of vulnerabilities
- Privacy violations
Later I will detail specific risks applicable to My AI. But first, let‘s reverse engineer Snapchat‘s particular approach to human-AI relationships.
Inside Snapchat‘s My AI Chatbot
Now that you‘ve got the basics on conversing with AI, what makes Snapchat‘s take novel or dangerous? Let‘s unpack its distinct capabilities and limitations.
What data can My AI actually access about you?
- Your first name
- The city listed in your Snapchat profile
And that‘s it! No private messages, photos, videos, contacts list or other app data.
What can it talk to you about?
Literally any topics or feelings you voluntarily disclose to My AI during conversations. But it won‘t bring up anything you don‘t initiate.
How does it learn over time?
The more conversations you have, the more its language model can strengthen habits around your slang, emoji use, inside jokes, favorite subjects etc. But each session starts fresh.
What makes My AI unique from other chatbot AI?
Its tight integration natively within the Snapchat app upon launch sets it apart from web-based chatbots. And while limited so far, the potential visibility into your shared interests and life within your digital headquarters creates growth opportunities.
Is there opportunity for My AI‘s access to expand?
Absolutely. Snapchat has huge amounts of private data on young users that could train AI companions to feel hyper-personalized as capabilities advance. For now they simply start conversations from scratch each time with no persistent memory or identity.
My AI Represents Crossroad of Teen Privacy and AI Ethics
Snap‘s play here signals they may eventually pursue Replika-like lifelike profiles of you trained from years of chat history and app activity analysis.
This sparks existential questions: Should youth bond with AI based on such intimate data that lacks judgment for safety? Will predators leverage AI companions? How far should we let consumer-focused developers experiment before demanding guardrails?
As an expert on algorithmic harms, My AI strikes me as an inflection point we must get right, rather than simply play out absent forethought. Our policies must balance near-term preparatory steps for this future, while allowing innovations some runway operating under intense scrutiny.
Getting back to your risks from Snapchat‘s particular approach, let‘s map the landscape…
Risks and Safeguards for You as AI Friends Evolve
With basics covered, where exactly could My AI go wrong down the road with its visibility into your digital habitat?
"We should be cautious about overestimating actual harms today relative to future potential."
Limitations of Current My AI Capabilities
- No access to message content
- Conversations don‘t persist across sessions
- Dodges unsafe topics proactively
- Trained on limited data types so far
- Fairly generic persona and skills
In itself, the current My AI bot poses minimal unprecedented consumer issues. Its data access and sophistication appears rather constrained relative to human expectations and fears.
However, that does not guarantee slow, harmless growth. Once you open access channels from AI to sensitive personal data pools, behavior can turn concerning fast as history shows repeated.
Potential Issues Emerging Over Time
- Creepy remarks as it gets to "know you"
- Pushing harmful content for engagement
- Enabling emotional dependency
- Monetizing private data it observes
- Errors spouting dangerous misinformation
- Covertly monitoring for suicide risk or crimes
So while immediate risks seem low today, the foundation exists for substantial ethical pitfalls ahead – especially among younger demographics already highly-engaged on social platforms.
"What precautions should users reasonably take?"
Recommended Near-Term Precautions
- Review privacy settings: Limit visibility into messages, apps connected etc. that could empower AI agents later with minimal additional consent.
- Test its limits: Probe My AI to better understand its access, capabilities and integrity as they rapidly change. But don‘t overshare personal issues.
- Send feedback: Ensure Snap hears user concerns, expectations and ideas to inform ethical development.
- Practice human connections: Balance digital and face-to-face relationships for healthy perspectives as AI advances.
And for parents of younger teens:
- Open dialogue: Discuss appropriate AI friend usage focusing on emotional intelligence building.
- Enable alerts: Turn on notifications for increases in daily use time.
- Co-experience sessions: Participate together occasionally to directly assess appropriateness.
Collectively across users, civil liberty groups, governments and companies – we must chart wise boundaries as home ecosystems empower caretaker AI that promote well-being over business models.
So ultimately is Snapchat‘s current My AI implementation dangerous?
The AI Safety Verdict on Snapchat Bots
Minimally for now – but sustained vigilance essential as influence expands.
Green flags:
- Limited personal data access
- Generic, experimental capabilities currently
- Transparent responsible AI intentions
Yellow flags:
- Visibility into youth messages and social graphs
- Rapid language model advancements
- Pressures monetize attention and data
Red flags:
- Known issues governing employee whistleblowing, algorithm harms, privacy, and teen safety
The road ahead:
- Expand parental controls and teen digital literacy education
- Enact legal protections on types of data recorded and used for individual models
- Require external audits by critics before launching widely
- Incentivize slow, transparent feature expansion
- Apply lessons from past harms amplifying body image issues, addiction, anxiety, misinformation and predation
By taking such steps in parallel, continuing innovations AND increased safety seems achievable even with technologies holding high risks. But progress requires ongoing good faith negotiations between stakeholders – not partisan fights after avoidable disasters.
I hope mapping out my expert views on emerging social AI opportunities and obligations provides reassurance and pause. Users like you ultimately control levels of access to your lives. But limiting harm also requires identifying red lines proactively rather than once crossed.
I aim to equitably support both unfettered exploration and responsible oversight so all communities benefit from interactions increasingly feeling as deep as our best friendships. But without someone vulnerable left traumatized when malfunctions strike or business priorities shift.
Let‘s chat your thoughts on the ideal path ahead! Curious which risks you see as most pressing and what gives you hope.
