Artificial Intimacy: How Generative AI Can Now Create Your Dream Girlfriend
Imagine being able to create your perfect partner – picking out items from a list to determine everything about them. You want red hair? You got it! Like a bubbly, fun personality? Check. Want them to be a biological, sentient being you can physically touch? Oh well, you can't have it all!
Artificial Intimacy: How Generative AI Can Now Create Your Dream Girlfriend
Adobe StockToday, you can choose to spend the wee hours talking sweet (or not-so-sweet) nothings to whoever you want, as long as you're happy that they're a digital personality and you're willing to pay for the experience.
One provider of such services is DreamGF, which makes the tantalizing promise that your virtual girlfriend is just a few clicks away.
This comes at a time when a growing number of influencers are discovering that creating a virtual avatar to talk to fans on their behalf can be a time-saving and lucrative side hustle.
So how does this work and perhaps more importantly, what are the implications for the future of relationships between real living, breathing human beings?
I decided to have a chat with two executives of DreamGF to try and find out.
AI-Generated Girlfriends
DreamGF allows users to design and then attempt to form relationships with girlfriends whose images and personalities are entirely constructed by generative AI.
Users can choose from a range of physical attributes, including hair length, ethnicity, age, and breast size. They can also choose from a much smaller set of personality attributes to decide whether she is a nympho, dominatrix, or nurse.
Once she's manifested in the digital universe, they can chat with her, including sexting her, and also ask her to send them pictures of an adult nature.
Although there is an option to select lesbian as a personality trait, it seems pretty clear that the service is designed for heterosexual men. However, I'm told that a DreamBF version is in the works.
Image generation is handled by the open-source Stable Diffusion engine. A clever innovation is that coding tricks have been used to overcome the fact that it's often difficult to create characters with a consistent appearance using AI image generators.
To handle the chat functionality, however, DreamGF has trained and deployed its own large language model. This was necessary because commonly used LLMs like OpenAI's GPT-4 don't allow adult content.
The technical wizardry doesn't stop there - it's even possible for users to receive voice messages from their AI girlfriends.
Although styled and marketed as a dating experience, it quickly becomes clear that the primary use case is to provide adult entertainment. This is a fact that DreamGF CEO Georgi Dimitrov and VP of Business Development Jeff Dillon were happy to admit when they joined me for a recent webinar discussion.
The Business Case
It's pretty clear that there's a business case here; after all, a lot of the Internet was built on porn and sex sells.
Dimitrov and Dillon both come from backgrounds in the adult industry, ranging from running OnlyFans agencies to payment processors, NFT projects, and electronic sex toy development.
With a wealth of cumulative insight into the direction of the adult entertainment industry, both believe that AI is the future.
However, the success of OnlyFans - built on enabling viewers to develop (seemingly) one-on-one relationships with content creators - is also central to the business model.
"You can be more transparent now and say, hey, you're chatting with the AI version of me," says Dillon. We'll come back to this later.
So far, things seem to be going well from a business point of view. Although only recently launched, the service has around 10,000 paying users and, perhaps more importantly, a retention rate of 80 percent. It also offers a free trial, and in total, over six million virtual girlfriends have been created, sharing around 20 million messages with their human partners.
Dillon says, "Obviously, there's a lot of people who are fascinated with the technology, and they just want to try it out to see how it works because it's interesting.
"But a lot of people actually stay and engage in the long run with their AI girlfriend."
Dimitri adds, "If you look at the market, there are many image generators, there are many text chats, and there is no solution that offers the full end-to-end experience of creating your girlfriend, chatting with her, and requesting content."
Social And Ethical Implications
Obviously, the growth in popularity of services based around enabling AI relationships and sex raises a great many ethical questions.
Dillon and Dimitrov say they are more comfortable facilitating relationships with AIs than with OnlyFans models and influencers due to the fact that many OnlyFans models lie about the fact that they are directly engaging with fans.
Dillon is happy to talk about that particular platform's open secret – "This kind of deceptive practice – where you think you're chatting with the model. But it's some guy in India.
"I was feeling at one point that essentially, we're lying … And the answer for me was AI."
This is clearly only the start of the ethical questions that arise due to the existence of such a service, though.
The platform has put measures in place to ensure it can't be used to generate illegal content, for example featuring underage characters. It also prevents users from creating virtual girlfriends based on real people. This is done by abstracting the image generation prompts away from the user through the site's user interface.
Dimitrov says, "We don't leave the people to write exactly what they want … because you're going to get a lot of celebrities … and children … this is something that we don't want to leave in the hands of the user."
The chat algorithms themselves are designed to detect and ban words related to child pornography.
"There's a lot of people who are interested in these kinds of things … and we are fighting with all the technology that we have. This is our safeguard," he says.
Other issues, such as whether there is a danger that users might begin to form relationships that are unhealthy or detrimental to their ability to form real-life relationships, aren't so easily solved with technology alone.
If men – particularly relatively young and immature men – become used to the idea of virtual girlfriends who are basically programmed to give them whatever they want, how will this affect the way they view and treat real women?
Dillon's answer to this is that new technology is often a double-edged sword, and while there may be negative consequences for some people, this has to be balanced with the potential for positive change.
He says, "It's like, 'Hey, I'm kind of used to this convenience that technology has given us.'
"They get used to talking to a girl that they can say 'I want a submissive girl' [and then] in real life, they develop a relationship and she's not so submissive and, you know, there's going to be some adjustment period for some people.
"But one of the cool things … it could be used for many different things. It could be used for relationship therapies … a lot of people are afraid to share things that are going on in real life … but talking to an AI person … I feel much more comfortable sharing my feelings because I'm just talking to a robot."
The Future Of Virtual Girlfriends
Whatever your view on the moral or ethical implications, I think it's clear that the AI sex and relationship industry has arrived, and, for better or worse, it isn't going to go away any time soon.
At DreamGF, the focus in the immediate future will be on digitizing real models to create hybrid girlfriends – who exist in real life as well as AI-generated avatars on a computer or smartphone screen.
Dillon says, "There's these models that have massive fan bases … for us to be able to create the AI version of Star X – so now she can point her fans to this AI version of herself, and they can interact with her right now … that's probably our big step going forward."
In fact, DreamGF is planning to go live with its first hybrid AI model in the next few weeks.
Beyond that, there's video. Currently, this is the holy grail for many generative AI platforms, but it's easy to foresee that a time will come in the relatively near future when it will be possible to have real-time video chats with AI girlfriends (and boyfriends).
And even further into the future? Well, then, there are robots.
Dimitrov says, "The most exciting thing … we're even creating an option to export your [virtual girlfriend] – the chat history and everything, the whole experience – and put it into something like a real-life robot.
"In Japan … they can create a real-looking robot with the features of a person … I think that's maybe ten or twenty years' time. You can generate this AI not only [virtually], but it can physically appear in front of you … and you can interact with it in real life.
"This feature, on the one hand, gets me excited, but on the other hand, quite scares me."
I don't think he's the only one!
You can click here to watch my conversation with DreamGF's Georgi Dimitrov and Jeff Dillon in full, where we go into more depth on some of the ethical issues around the AI sex and relationships industry and what the future might hold.
You can read more about these topics in my books, The Future Internet: How the Metaverse, Web 3.0, and Blockchain Will Transform Business and Society, Future Skills: The 20 Skills And Competencies Everyone Needs To Succeed In A Digital World and Business Trends in Practice, which won the 2022 Business Book of the Year award. And don't forget to subscribe to my newsletter and follow me on X (Twitter), LinkedIn, and YouTube for more on the future trends in business and technology.
AI And The Economy: Customer Service Jobs Will Be Cut, Quality Improved
Workers at a call center (AP Photo/David A. Lieb)
Associated PressCustomer service has greatly benefited from artificial intelligence already, though many customers and company staff members have also been frustrated. AI has had problems, but better AI, better management and better training will enable better customer service at lower cost.
Previous articles in this series, AI and the Economy, have addressed broad economic themes and a few specific sectors. The customer service function provides the lowest hanging, fattest fruit in the whole orchard.
"Customer support agents using an AI tool to guide their conversations saw a nearly 14% increase in productivity," according to an academic study. The greatest gain was for less experienced agents, up 35%, though the most experienced agents had no improvement on average.
Today businesses across the United States employ nearly three million customer service representatives at an annual salary of over $40,000. A little productivity boost here will pay a large dividend.
One company's AI was able to cut 36 seconds off average call time simply by routing calls to the appropriate department, according to a Wall Street Journal article. That may not sound like much, until multiplied by thousands of calls. Customers are very sensitive to time wasted on a call; that is, time spent waiting to get to someone who can help, or time for that person to pull up relevant information.
AI cannot only answer questions about how to help a customer, it can also detect the caller's mood , such as angry, confused, frustrated or happy. Sentiment analysis can help in the moment and can also be a tool for development of better practices, both by the AI and by human customer service representatives.
Records of past interactions with a customer are valuable. Older AI products are good at converting an audio recording to a transcript, and newer AI is very good at summarizing those transcripts. The summary will be useful for follow-up calls as well as improving future support.
Customer service is not just about call centers. Hands-on work can be recorded on video and then summarized by AI. And videos are now recorded by some service technicians. I've seen them for my car repair and for HVAC service. A summary can help both the customer and the next technician to work on the project.
Challenges have arisen from the use of AI in customer service. Some experienced agents believe that the AI's suggestions are not as good as theirs. And some companies have given conflicting orders about when to follow the AI and when to use human judgment. It's always bad management to give a person responsibility without the authority to accomplish the task. Clarity about the human's role is crucial to success.
Another challenge comes from large language models' tendency to hallucinate, which is industry jargon for making up falsehoods. Unfortunately, the AI sound quite confident it is correct even when it is egregiously incorrect. Systems should be designed to consider the consequences of an hallucination. A call being routed to the wrong department is not a big deal. Money being transferred to the wrong account is a very big deal.
The best practices, as we currently understand them, begin with setting up the AI to do a good job. In some cases fine tuning the model is appropriate. That takes the large language model and re-estimates its parameters to achieve a particular end, such as understanding the jargon of an industry or product. It is a somewhat expensive task, even though much cheaper than developing the model from scratch. Less expensive alternatives, such as a tool called LoRA, seem to work almost as well. Prompt engineering is a cheaper alternative that often gets suitable results. For example, an AI can be told to always make its answer consistent with the company's policy book, its owners' manuals and its troubleshooting guides, which are made available in the prompt.
Data security is a danger that must be addressed. Off the shelf, large language models can use the information submitted in a prompt. For example, if the AI is asked a question that includes a caller's name and bank account balance, that information goes to the AI company. Most of the specialized systems install walls to keep that information confidential. It's not hard to set up, and businesses should make sure information is protected.
Another good practice for automated call systems is an easy path to talking to a human, according to wise advice from helpwise.Io.
The academic study of call center productivity mentioned above raises two interesting questions. First, will the less experienced agents come up to speed faster for having the AI help them, or will they not develop their skills the way that today's experienced agents did? Second, will the AI improve enough in the coming years to help even the more experienced agents? There is no doubt that large language models such as ChatGPT, Bard and Claude will improve, but it's not a sure thing that they will ever surpass top-notch customer service agents, especially with the most difficult questions.
AI is here to stay. It can be implemented with priority given to improving customer service, or with priority to minimizing costs for a given level of service. That's a critical management decision.
The number of customer service representatives, about three million now in the U.S., will certainly decline in the coming years. And in most cases, customer service will improve. The early shrinkage of jobs may be accomplished through normal attrition, but layoffs are not far away. The best agents will be retained to handle difficult issues, but other people in this sector should start thinking about career alternatives.
AI Should Be Decentralized, But How?
The intersection of Web3 and artificial intelligence (AI), specifically in the form of generative AI, has become one of the hottest topics of debate within the crypto community. After all, generative AI is revolutionizing all areas of traditional software stacks, and Web3 is no exception. Given that decentralization is the core value proposition of Web3, many of the emergent Web3-generative-AI projects and scenarios project some form of decentralized generative AI value proposition.
Jesus Rodriguez is the CEO of IntoTheBlock.
In Web3, we have a long history of looking at every domain through a decentralization lens, but the reality is that not all domains can benefit from decentralization, and for every domain, there is a spectrum of decentralization scenarios. Breaking down that idea from a first principles standpoint leads us to two key questions:
Does generative AI deserve to be decentralized?
Why hasn't decentralized AI worked at scale before, and what's different with generative AI?
What are the different dimensions of decentralization in generative AI?
These questions are far from trivial, and each one can spark passionate debates. However, I believe that thinking through these questions is essential to develop a comprehensive thesis about the opportunities and challenges at the intersection of Web3 and generative AI.
Does AI Deserve to be Decentralized?
The philosophical case for decentralizing AI is simple. AI is digital knowledge, and knowledge might be the number one construct of the digital world that deserves to be decentralized. Throughout the history of Web3, we have made many attempts to decentralize things that work extremely well in a centralized architecture, and where decentralization didn't provide obvious benefits. Knowledge is not one of the natural candidates for decentralization from both the technical and economic standpoint.
The level of control being accumulated by the big AI providers is creating a massive gap with the rest of the competition to the point that it is becoming scary. AI does not evolve linearly or even exponentially; it follows a multi-exponential curve.
GPT-4 represents a massive improvement over GPT 3.5 across many dimensions, and that trajectory is likely to continue. At some point, it becomes unfeasible to try to compete with centralized AI providers. A well-designed decentralized network model could enable an ecosystem in which different parties collaborate to improve the quality of models, which enables democratic access to knowledge and sharing of the benefits.
Transparency is the second factor that can be considered when evaluating the merits of decentralization in AI. Foundation model architectures involve millions of interconnected neurons across several layers, making it impractical to understand using traditional monitoring practices. Nobody really understands what happens inside GPT-4, and OpenAI has no incentives to be more transparent in that area. Decentralized AI networks could enable open testing benchmarks and guardrails that provide visibility into the functioning of Foundation Models without requiring trust in a specific provider.
Why Hasn't Decentralized AI Worked Until Now?
If the case for decentralized AI is so clear, then why haven't we seen any successful attempts in this area? After all, decentralized AI is not a new idea, and many of its principles date back to the early 1990s. Without getting into technicalities, the main reason for the lack of success of decentralized AI approaches is that the value proposition was questionable at best.
Before large foundation models came into the scene, the dominant architecture paradigm was different forms of supervised learning that required highly curated and labeled datasets, which resided mostly within corporate boundaries. Additionally, the models were small enough to be easily interpretable using mainstream tools. Finally, the case for control was also very weak, as no models were strong enough to cause any level of concern.
In a somewhat paradoxical twist, the prominence of large-scale generative AI and foundation models in a centralized manner helped make the case for decentralized AI viable for the first time in history.
Now that we understand that AI deserves to be decentralized and that this time is somewhat different from previous attempts, we can start thinking about which specific elements require decentralization.
The Dimensions of Decentralization in AI
When it comes to generative AI, there is no single approach to decentralization. Instead, decentralization should be considered in the context of the different phases of the lifecycle of foundation models. Here are three main stages in the operational lifespan of foundation models that are relevant to decentralization:
Pre-training is the stage in which a model is trained on large volumes of unlabeled and labeled datasets.
Fine-tuning, which is typically optional, is the phase in which a model is "retrained" on domain-specific datasets to optimize its performance on different tasks.
Inference is the stage in which a model outputs predictions based on specific inputs.
Throughout these three phases, there are different dimensions that are good candidates for decentralization.
The Compute Decentralization Dimension
Decentralized computing can be incredibly relevant during pre-training and finetuning and may be less relevant during inference. Foundation models notoriously require large cycles of GPU compute, which are typically executed in centralized data centers. The notion of a decentralized GPU compute network in which different parties can supply compute for the pre-training and finetuning of models could help remove the control that large cloud providers have over the creation of foundation models.
The Data Decentralization Dimension
Data decentralization could play an incredibly important role during the pre-training and fine-tuning phases. Currently, there is very little transparency around the concrete composition of datasets used to pretrain and finetune foundation models. A decentralized data network could incentivize different parties to supply datasets with appropriate disclosures and track their usage in pretraining and fine-tuning foundation models.
The Optimization Decentralization Dimension
Many phases during the lifecycle of foundation models require validations, often in the form of human intervention. Notably, techniques such as reinforcement learning with human feedback (RLHF) enable the transition from GPT-3 to ChatGPT by having humans validate the outputs of the model to provide better alignment with human interests. This level of validation is particularly relevant during the fine-tuning phases, and currently, there is very little transparency around it. A decentralized network of human and AI validators that perform specific tasks, whose results are immediately traceable, could be a significant improvement in this area.
The Evaluation Decentralization Dimension
If I were to ask you to select the best language model for a specific task, you would have to guess the answer. AI benchmarks are fundamentally broken, there is very little transparency around them, and they require quite a bit of trust in the parties who created them. Decentralizing the evaluation of foundation models for different tasks is an incredibly important task to increase transparency in the space. This dimension is particularly relevant during the inference phase.
The Model Execution Decentralization Dimension
Finally, the most obvious area of decentralization. Using foundation models today requires trust in infrastructures controlled by a centralized party. Providing a network in which inference workloads can be distributed across different parties is quite an interesting challenge that can bring a tremendous amount of value to the adoption of foundation models.
The right way to do AI
Foundation models propelled AI to mainstream adoption and also accelerated all the challenges that come with the rapidly increasing capabilities of these models. Among these challenges, the case for decentralization has never been stronger.
Digital knowledge deserves to be decentralized across all its dimensions: data, compute, validation, optimization, execution. No centralized entity deserves to have that much power over the future of intelligence. The case for decentralized AI is clear, but the technical challenges are tremendous. Decentralizing AI is going to require more than one technical breakthrough, but the goal is certainly achievable. In the era of foundation models, decentralized AI is the right way to approach AI.
